The 1987-89 Locust plague in Mali: Evidences of the heterogeneous impact of income shocks on education outcomes †
∗
‡
Philippe De Vreyer, Nathalie Guilbert and Sandrine Mesplé-Somps
§
April 16, 2012
Abstract This paper estimates the long run impact of a large income shock, by exploiting the regional variation of the 1987-1989 locust invasion in Mali. Using exhaustive Population Census data, we construct birth cohorts of individuals and compare those born and living in the years and villages aected by locust plagues with other cohorts. We assert that earliest childhood exposure to income shock had a larger negative eects on the probability to go to school than later childhood exposure. Indeed, the proportion of boys born during the shock and who later enrolled at school is reduced by 6.1% if they lived in a community invaded by locusts, and by 3.5% for girls. This impact goes up to 7.5% for boys and 5% for girls living in rural areas. Children from invaded rural localities have completed up to one lower grade than if they had lived in another community. Educational attainment of all cohorts of girls living in rural areas and aged less than 11 years old during the plague has also been aected. Controlling for variations in the rainfall level or the potentially selective migration behavior of individuals do not dampen our results. Our results are also robust to dierent variations of the cut-o cohort.
Keywords: Education, Shocks, Mali, Locust. JEL classication codes: I21, O12, O55.
We thank the Malian Institute of Statistics (INSTAT) for giving us data access and the CEPREMAP for providing nancial support to this research. We also thank participants at seminars and conferences in Paris and Oxford. The usual disclaimer applies. † Université Paris Dauphine, LEDA, and IRD, UMR 225 DIAL. E-mail:
[email protected] ‡ Université Paris Dauphine, LEDA, and IRD, UMR 225 DIAL. E-mail:
[email protected] § Institut de Recherche pour le Développement (IRD), UMR 225 DIAL and Université Paris Dauphine, LEDa. E-mail:
[email protected] ∗
"This is what the LORD, the God of the Hebrews, says: 'How long will you refuse to humble yourself before me? Let my people go, so that they may worship me. If you refuse to let them go, I will bring locusts into your country tomorrow. They will cover the face of the ground so that it cannot be seen. They will devour what little you have left after the hail, including every tree that is growing in your elds. They will ll your houses and those of all your ocials and all the Egyptians, something neither your fathers nor your forefathers have ever seen from the day they settled in this land till now." EXODUS, 10:3-6
1
Introduction
The consequences of shocks undergone during early-life on human capital formation and on the well-being of adults have attracted considerable academic and policy interest. If economic shocks reduce child human capital investment, they may transmit poverty between generations and maintain people in poverty for a long time. Numerous shocks can impact human capital investment of children living in low income countries ranging from idiosyncratic shocks due, for instance, to job loss or death of adult family members to large macroeconomic shocks, such as those caused by macroeconomic crisis or natural disasters.
Recent papers have documented long-lasting eects of such shocks on adult outcomes such as educational attainment, socio-economic status, income, cognitive ability, disease, height or life expectancy.
They conrm the fetal origins hypothesis (Barker, 1992):
poor
environmental conditions during in-utero and early childhood that induce shocks to nutrition can have permanent eects on physiology and adverse consequences on later life outcomes.
et al., 2007; Currie and Moretti, 2007) as well as in developing countries (Dercon, 2004; Case et al., 2005; Alderman et al., 2006; Maccini and Yang, 2009; Gorgens et al. (2011); Leon, 2009; Grimard Evidence has been gathered in developed countries (Almond, 2006; Banerjee
and Laszlo, 2010). Ferreira and Schady (2009) provide a literature review on the impact of aggregate economic shocks on child schooling and health, whereas Alderman (2011) produces a synthesis of recent works on the impacts of shocks in early childhood development.
Establishing a causality between conditions during early life and outcomes later in life is the main concern of most of the recent research papers. A promising way to identify any causal link is to analyze the consequence of exogenous shocks, like pandemics, extreme drought
2
or civil war and exploiting the variation in the temporal and geographical incidence of these exogenous shocks. Almond (2006) uses the 1918 inuenza pandemic as a natural experiment for testing the long term eects of in-utero inuenza exposure on several American adult outcomes. He estimates that children whose mother has been infected had a probability to graduate from high school up to 15% lower than other children, men wages were reduced by 5 to 9% and the probability of being poor increased by 15% for aected cohorts. Banerjee
et al.
(2007) identify the impact of Phylloxera, an insect that attacks the roots of grape wine and destroyed 40% of French vineyards between 1863 and 1890, on height and health outcomes of young male adults.
They estimate that children of wine-growing families born during
Phylloxera crisis were 0.6 to 0.9 centimeters shorter than others by age 20. Gorgens
et al.
(2011) estimate the long run impact of the China's Great Famine on survivor health outcomes and exploit a source of variation in the regional intensity of food shortage derived from an institutional determinant of the Great Famine. Controlling for selection, they found that rural famine survivors who were exposed to shortages in the rst 5 years of life are stunted between 1 and 2 cm. They also measure the selection eects and estimate that height-related selection has increased the average height of rural women survivors by about 2 cm. Leon (2009) and Grimard and Laszlo (2010) use the variation in the incidence of civil conict in Peru from 1993 to 2007 to analyze the impacts of such a conict on educational attainment and health outcomes. They show that cohorts of women in-utero during the conict are smaller than the other ones. Maccini and Yang (2008) examine less extreme and unusual early-life conditions,
i.e.
rainfall shocks in Indonesia, on health, educational and labor outcomes of adults. The
authors report striking results for women: those born in places experiencing a 20% higher rainfall than normal at their time of birth are 0.57 centimeters taller, 3.8% less likely to report poor or very poor health status, complete 0.22 more grades of schooling, and live in households that score 0.12 standard deviations higher on an asset index. Finally, Alderman
al.
et
(2006) exploit civil war as well as drought shocks to identify the long term consequences of
early childhood malnutrition on schooling in Zimbabwe. They show that children that were stunted at pre-school age were also 3.4 cm smaller young adults, started school 6 months later and completed less grades of schooling (0.85 grades) than other children.
In this paper, we consider the eects of a natural disaster that has made a lasting impression in the mind of generations of people: desert locust invasions. Surprisingly very little is known about the impacts of such a natural disaster, though it occurs regularly in Africa, the Middle East and South-West Asia and concerns a total of 65 countries.
This
maybe due to the lack of adequate data and to the fact that locust swarms are more likely
3
when rainfalls are high, so that their impact is mitigated by the higher crop yields that come with good rains.
However, even if at the macroeconomic level the impact of locust
invasions appears small, at the household level it can be very high for farmers which crops have been eaten. We estimate the long run impact of the 1987-1989 locust invasion in Mali on educational attainment outcomes using its regional variation inside the Malian territory. As the 1987-1989 locust invasion induced large crop shortages in aected regions but not national famine, we are able to identify non aected villages. Using the 1998 exhaustive Population Census data, we construct birth cohorts of individuals and compare those born and living in the years and villages aected by locust plagues with other older cohorts whose education was not impacted by the plague, while controlling for rainfall variations, using historical climate data.
Beyond being the rst paper to estimate the long term impact of locust invasion, the main contribution of this study is to oer some insight on the likely eects of local or idiosyncratic shocks to which households in developing countries are frequently submitted, but that are dicult to observe in surveys. Locust invasions, because they strike randomly and are of a limited scope, but at the same time concern a large enough number of people, can be used as a natural experiment to analyze households ability to deal with the impact of such shocks.
We nd that children whose household has been exposed to locust invasion while they were in age of school admission or younger have a lower probability of going to school than other children. Indeed, the proportion of boys born during the shock and who later enrolled at school is reduced by 6% and by 3.5% for girls. Distinguishing among residential areas, we nd stronger impacts on children from rural areas, with respective decreases in school enrollment of 7.5% and 5%. On the other hand, no impact is found on the educational outcomes of children living in urban localities. Regarding educational attainment, we nd a negative impact on the number of years of education and on the probability to achieve primary level for children in age of starting school at the time of plague.
The shock has impacted more deeply and
widely girls educational attainment than that of boys, with respective largest eects of one and 0.44 lower grade attended for children living in invaded communities. Among enrolled children, more than 10% have not achieved their primary level if they experienced the shock at the time of school admission. Our main results are found for the resident population sample i.e people who have never moved from their birth place. Further estimations conrm that holding account of the migrant population does not alter previous ndings. Moreover, we check wether our ndings are driven by the cohort cut-o point of the sample, and nd that this is not the case.
4
Our paper is organized as follows. Section 2 discusses the causes and consequences of locust invasions. Section 3 presents the empirical strategy and the data. Section 4 presents the results and section 5 some robustness checks. Finally section 6 concludes.
5
2
Locust invasions: origins and consequences
Mali is a large (1,242,000 square kilometers), sparsely populated (13 millions inhabitants in 2009) and low income (GDP per capita was $691 in 2009) country between the 10th and the 25th parallel. As such a large part of its territory is located in the Saharan part of Africa, a region threatened by drought and desertication that can hardly be used for agriculture. Poverty is high (headcount index was 61% in 2001 at the $1,25/day/capita absolute poverty line) and life expectancy very low (48 years in 2008), together with the literacy rate (26% in 2006, but in rapid progression, since it was only 19% in 1998). Malnutrition remains at a very high level: in 2006, 38,5% children under ve had a height for age Z-score more than two standard deviations below the median for the international reference population. Agriculture
1
employs about 40% of the active population and brings 37% of GDP (in 2007).
The country
is very much submitted to natural and other external shocks due to its high dependence upon agriculture and the concentration of its exports on three commodities (gold, cotton and livestock).
Among these shocks, locust invasions maybe the less frequent, but one of the most impressive, as exemplied by the citation at the top of this paper.
The locust plague is
the curse of good rains as it generally comes when precipitations are higher than average. The Desert Locusts (DL) live as harmless solitarious individuals in areas that are not, or only minimally, used for agriculture and have average annual precipitation of no more than 200 mm. These areas (called recession area) are distributed across several Sahel countries (see gure 1). When environmental conditions become favorable, mainly adequate, evenly distributed rainfall over a period of several years (Duranton and Lecoq, 1990), mass reproduction takes place. The increasing density then changes the insect's behavior and stimulates a gregarious phase which results in swarms of billions of insects. Those bands are able to migrate very long distances outside the recession area and pose a threat on agricultural productions in 65 countries of Africa, Middle East and South-West Asia, covering 29 millions square kilometers. Swarm size can be very large, varying between less than one square kilometer to several hundred square kilometers. Since there can be at least 40 millions and sometimes as many as 80 millions locust adults in each square kilometer of swarm and since a Desert Locust adult can consume roughly its own weight in fresh food per day, that is about two grams every day, one gets an idea of the amount of damage an average size swarm can indulge on a rural
1 Source:
World Development Indicators, World Bank 2009. The share of the active population employed in the agricultural sector is extracted from national accounts. It seems to be underestimated compared to the 1998 Population Census data that estimates this share around 81%. 6
locality.
A one square kilometer swarm, with 60 million insects can eat about 120 tons of
food, that is enough to feed 2500 people during about 4 months.
Fortunately, the Desert
Locust diet is not limited to the fruits, cereals and vegetables human being eat, so that the damage might not be as bad as could be feared. Latchininsky and Launois-Luong (1997), in a monographic study of Desert Locusts in Central Asia and Transcaucasia, give a detailed list of more than 150 botanical species of all kinds. They mention other studies reporting as much as 400 species.
[insert gure 1 about here]
In the absence of preventive control, waves of locust invasions can succeed with a high frequency and last for as many as 22 years. From 1860 to 2004, a total of nine invasions have taken place: 1860-1867, 1869-1881, 1888-1910, 1912-1919, 1926-1935, 1940-1947, 1949-1962, 1987-1989 and 2003-2004 (Lecoq, 2004). The costs of these invasions is not easy to estimate precisely, mainly because of lack of adequate data, and because invasions occur when rainfall are higher than average. Thus, in Mali, the 1987-1989 invasion did not result in major crop losses, at a macroeconomic level. On the contrary, in 1988, which is the year with the highest number of areas reporting locust swarms, yields for cereals were also at their highest (see gure 2). According to Thomson and Miers (2002), even when net damage is reported it does not go beyond 2 to 5% of total production. In face of this, a debate has emerged about the opportunity to prevent and control the locust plague and how this should be done. Prevention supposes a close monitoring of the recession areas. As these are remote, sparsely populated areas, such control is costly to enforce. If successful, locust activity can be controlled before it threatens crop production. The second possibility is to wait until swarms have developed and are numerous, at which point a greater impact can be obtained, because of the greater density of locusts. At this point the massive chemical spraying of large areas remains the preferred weapon, in spite of its cost (300 millions euros spent in 1988, Lecoq 2004) and of its negative impact on the environment and on the health of farmers. Joe (1997) attempts to present a cost-benet analysis of Desert Locust Control. According to his results, preventive campaigns do not bring enough benets in regard of their cost. The main argument in support of this conclusion being that even in the worst case scenario of massive destructions by swarms the cost of the lost productions barely amounts to that of preventive control. Moreover, as locust swarms cross borders, the benets of one country's eorts to control locusts can be annihilated if neighboring countries do not invest at the same level. These considerations militate in favor of an insurance scheme, that would protect farmers against the risk of locust swarms, without incurring the monetary, health and environmental costs of chemical warfare.
7
The need for Desert Locust Control or for the compensation of invaded farmers can only be assessed through a better knowledge of the incurred costs. Indeed, even if low at the macroeconomic level, the impact of locust invasions can be high at a local or regional level. Swarms invasions are local by nature and there could be severely aected regions, or villages, in which major problems have been caused by the destruction of all or part of the harvest. But diculties in this case do not come from aggregate shortages, but rather from distribution problems.
This is conrmed by the Famine Early Warning System for Mali which reports
that food shortages experienced during those years were caused not by pests, but rather by unequal distribution of food (Herok and Krall, 1995). Thomson and Miers (2002) have used eld interviews to evaluate the impacts of swarms invasions in Mauritania and Eritrea. Peasants in both countries mention the lack of water as the rst impediment to their farming activities. When talking about pests, farmers in Mauritania appeared more worried by the small, but regular, losses incurred due to birds, caterpillars, termites, ticks, rats, plant louse,
"when the subject of locusts was raised, it became clear that these are regarded as an altogether dierent type of hazard, a periodic shock causing total destruction to an extent that is incomparable with the regular damage of other pests. A locust plague will eat an entire harvest and will leave no pasture for animals to graze. Most respondents (...) used vocabulary such as "catastrophe", "crisis", "disaster", reecting the severity of the destruction and placing it on the same level as the last major drought. There is a saying that if a locust lands on a stone it will eat the stone" squirrels, snakes, scorpions, jackals and monkeys. However,
(Thomson and Miers, page 11). These interviews conrm that farmers that lost part or all of their harvest due to locusts can be severely hit. In this paper, we shall look at the long term impacts of such shocks, focusing on the human capital building of young children.
[insert gure 2 about here]
The expected consequences of locust invasions at the household level are not completely straightforward. Theoretically locust invasions can have negative consequences for the entire population if a signicant proportion of the available food is destroyed by the swarms and if it results in increasing ination. But, as we have seen, it does not seem likely. Hence, the impact sign and size will depend mainly upon the household location and activity on the labor market. Farmers in invaded villages are expected to be more concerned than teachers in non invaded villages for instance. Locally, in invaded villages, some households could prot from locusts, but it will depend on the markets village integration. If access to the food market is easy, then the destruction of harvests in a given village should not result in an increasing price of food. Only the farmers whose production has been destroyed should suer through
8
a reduction of their income. Those who exert their activity in the transport or commercial sector could benet from the invasion, since the demand for their services increases. In case the village has no access to the food market, the local price of food would increase following the invasion. Household with low income and with low mobility would then suer from the price increase even if they are not farmers. Besides farmers, breeders are another category at risk since locusts eat the same food as their cattle, but the size of the impact on this category will also depend upon their ability to access outside markets.
There is also the possibility
that the food destruction may be partly compensated by the increasing availability of protein that is brought by locust swarms. Indeed, locusts can be stir-fried, boiled or roasted and in many countries people eat locusts, particularly during outbreaks. However this can only be done when the swarms are not sprayed by chemicals. As concerns our outcome variable, educational enrollment or attainment, it could be impacted by locust swarms in several ways. First of all, if locust invasions result in lack of food, education of young children could be impacted because of a deteriorated nutritional status. Young children suering from a reduced diet maybe stunted or wasted, which could have a negative impact on their cognitive capacities. If invasion occurs during the in-utero life of the child, it could have long lasting eects on its health if the pregnant mother's health or nutritional status is impacted (Barker, 1992).
Second, the reduced income impact that
swarms can have on the household, could induce the poorest of these households to withdraw their children from school or to delay their school enrollment, in order to smooth consumption (Jacoby and Skouas, 1997).
3
Empirical strategy and data
3.1 Empirical strategy We assimilate locust invasions to a "treatment" administered to the invaded villages.
The
eect of this treatment is estimated using a dierence in dierence estimator. The fact that locust invasions have no observable impacts at the macroeconomic level provides us with an appropriate setting for evaluating their impact at the local level.
Impact evaluation is
based on the comparison of outcomes between invaded (so-called treated) and non invaded (untreated) areas and between potentially impacted and non impacted cohorts.
If locust
invasions have non negligible macroeconomic impacts, then the comparison between treated and untreated units will tend to under-estimate their impact, as non invaded areas could be contaminated through market price eects. The fact that global food availability does
9
not decrease signicantly during invasion years, guarantees that non invaded areas are not aected by the reduction in farms yields that occur in invaded areas.
Let
Scv
be a measure of educational investment (eg. enrollment) or outcome (eg. grade)
for people born in year
c
in village
v.
invaded by locusts and
C
the birth date of the observed individuals. The basic regression for
Let
Tv
be a dummy that equals 1 if village
v
has been
evaluating the impact of locust invasions on educational investment or outcome of cohort in village
v
is written:
Scv = α + βc .1{C=c} + γ.Tv + δc .1{C=c} .Tv + εcv where
c
(1)
δc measures the impact of the locust invasion on cohort c, γ accounts for xed dierences
between treated and untreated villages and
βc for dierences between cohorts that are common
to all villages. Treatment impact is captured by the interaction between the treatment dummy at the village level and birth cohorts.
One important feature for our concern is that locust invasions are more likely when rainfalls have been high for many years. This does not necessarily mean that villages that have been attacked by locusts have themselves beneted from high rains, because the breeding areas in which locust reproduce are not the same as the invasion areas. As concerns Mali for instance, this means that locust swarms form in the Saharan part of the country, but that harvests are more likely to be destroyed in the Sudanese-Saharan part of the country. Thus, though rainfall levels in the recession area are positively correlated with the probability of insects mass reproduction and swarms formation, there is no direct association between rainfall levels in a given village and the probability of a locust invasion in that village. However, when rainfall levels are higher than average in the Saharan part of Mali, there is a good chance that it will be also the case in the southern part of the country. For this reason we complete the model and control for precipitation levels around the birth date and the date of schooling of observed individuals in order to make sure that we do not confound the eects of rainfalls with those of locusts. Note that rainfall levels vary with geographical areas and cohorts. We also add a village xed eect in order to account for xed dierences between villages in the availability of schools and other relevant infrastructure.
Scv = α + βc .1{C=c} + δc .1{C=c} .Tv +
PL
l=1 (η−l .Rcv−l
+ η+l .Rcv+l )
(2)
+η.Rcv + µv + εcv where
Rt
is the measure of precipitations in year
t.
The xed eect model does not allow the
identication of the impact of xed dierences between treated and untreated villages. But
10
it remains possible to identify the treatment eect.
Though we observe the outcome variable for each inhabitant in the treated and untreated villages, the dependent variable in the model is the village average of this variable for each birth cohort. This choice is dictated by the fact that the treatment variable, together with other covariates, are observed at the village level and our choice of individual level variables is very restricted. Moreover, working with individual observations has it own disadvantages as one should hold account of the correlation of residuals between inhabitants of the same village. On the other hand, the use of averages introduces heteroskedasticity, since the number of inhabitants over which averages are computed varies from one village to another. In order to hold account of this heteroskedasticity we employ robust estimates of the variance-covariance matrix.
3.2 Empirical strategy Educational variables We construct a panel of birth cohorts using the exhaustive 1998 Population Census of Mali. The Malian 1998 Population Census data give information on the place and duration of residence, the age and the place of birth for each individual. The place of residence is known at the locality level (there are around 10,000 localities in Mali) whereas the place of birth is collected at the
cercle
level (50
cercles ).
We then rst restrict our sample to individuals
that never moved from their place. This could lead to an under-estimation of the impact of locust invasions if migration is more likely after a locust shock. On the other hand, we might under-estimate the impact of the shock if the migrant population leaving from the locusts impacted areas is signicantly more educated than the one leaving from the non impacted areas.
In the robustness checks section we undertake simulations to reallocate migrants in
the villages of their birth
cercle
proportionally to the village size and discuss to what extent
migration impacts our results.
We limit the sample to individuals from 33 to 7 years-old in 1998, that is to say individuals born from 1965 to 1991. For the sake of comparability, we exclude from the control group Bamako, the capital city that concentrates a huge part of the urban population of the country. As Mali is a very poor country with a very low rate of literacy and inecient birth certicate administration, individuals do not declare their date of birth but simply their age. This lack of precise data on birth date rst prevents us identifying exactly when individuals
11
have been aected by locust invasions, and, second, does not make possible to know if they were born during the dry or wet seasons that could inuence the rainfall impact on educational attainment (Maccini and Yang, 2009).
Table 1 gives the number of villages per cohort in the treatment group, control group, as well as the average number of individuals per cohort and group. It can be seen rst that, due to mortality, the oldest cohorts include less people than the youngest ones.
Second,
due to errors in the declaration of age and approximations around 10, 15, 20, etc.
years
old, cohorts 1988, 1983, 1978, 1973 and 1968 are more numerous than the cohorts close to them. For instance, the average number of 25 years old people (cohort 1973) per locality is 16 individuals compared to 9 individuals for the 1972 or 1974 cohorts. Nevertheless, cohorts 1990 to 1986 have been potentially aected by the 1987-89 locust plagues while in-utero and/or during early childhood, whereas children born between 1985 to 1976 were at the age of primary schooling during the 1987-89 locust invasions.
[ insert table 1 about here]
To measure educational attainment, we extract three variables: the enrollment rate (the proportion of individuals that have been at school), the number of classes attended at the primary school level by people attending school and the proportion of individuals that have achieved the primary level (among people that attended primary school). All these outcomes are computed for girls and boys separately. The graphs below (gures 4 and 5) plot the means of the three educational variables by cohort (born from 1965 to 1991) for all villages included in this analysis and separately for villages aected and not aected by locust invasions. As can be seen, the educational level of the cohorts born before 1982 is very low. Enrollment rates at the primary level started to increase only for cohorts born after 1982. Within ve years, it has doubled for boys and almost tripled for girls. In fact, Diara
et al.
(2001) reports a "non linear evolution" of gross
enrollment rate in Mali since independence, mainly due to lack of investment. First, it has increased rapidly during the 1960-1970s, then slowed down until decreasing during the 1980s before improving again during the 1990s until now.
This is illustrated by the breakpoint
occurring at cohort 1983, i.e. the cohort in age to enter school in 1990, on the enrollment rate graph.
Nevertheless enrollment rates are at best equal to 25% for boys and 16% for
girls at the middle of the 1990s (people born between 1986 and 91).
The boys enrollment
rate is approximately twice that of girls which mirrors the gender gap reality observed in the country. Indeed in Mali, as in many other developing countries, males are fully responsible of
12
their family material needs, and are in charge of providing income; therefore their education is considered more of a priority than that of girls. Moreover, some religious and traditional values, like early wedding and the gender allocation of domestic chores, do not promote girls school enrollment and attainment but keep them mainly in charge of household activities (Soumare, 1994; Diarra and Lange, 2000). Hence, in times of economic diculties, we suspect girls education to be more aected than that of their "brothers", either because priority in food allocation would be given to boys, leading to girls deteriorated cognitive capacities, or because girls manpower is requested to increase the earning capacities of the household. An other important feature is that, whatever the cohort of birth, less than 40% of people that attended primary school have achieved the Primary level (see the third graphs of gures
2
4 and 5).
[insert gure 4 and gure 5 about here]
The three education variables follow similar trends in locusts aected and non aected areas, for boys and girls. But a sizable divergence emerges between locust aected and non aected areas from cohort 1983: locusts aected localities experiment a much lower increase in enrollment rates. The gap between the two trends started for children aged 5 or 6 during the shock and keeps increasing for younger children. This means that children born in 1983 or after faced a very dierent educational environment than older cohorts. This could be a problem for our analysis if the dierence between the two groups of cohorts does not follow similar patterns among invaded and non invaded villages. Our village xed eects strategy prevents our results from being biased by a signicant dierent xed level of infrastructures between invaded and non invaded villages, but it cannot capture the potential dierentiated dynamics between the two areas.
Indeed, if the dynamics are dierent, this could partly
explain the educational gap between aected and non aected localities starting from cohort 1983. Further investigation on this matter is implemented in the robustness check section.
Locust localization and rainfall data The information on locust swarms localization is extracted from the FAO's Desert Locust Bulletins, produced by the Desert Locust Information Service (DLIS) and publicly available.
3
In each Bulletin, there are detailed information on locust swarms identication and
localization followed by forecasts. During periods of increased locust activity, bulletins are
2 Since
in Mali school starts at seven and the primary level is composed of six grades, only cohorts born before 1985 could have achieved the primary level in 1998 3 http://www.fao.org/ag/locusts/en/archives/archive/index.html 13
supplemented with alerts and updates. We code each Malian locality listed by these bulletins as having been aected by locust swarms between June 1987 to June 1989. Figure 3 places the 979 villages identied. The locust invasion spreads over an area on the middle of Mali that stretches from the East border to the West border of the country.
Some areas seem
particularly aected by locust swarms whereas others much less. Unfortunately, we cannot assert that these dierences are entirely due to locust invasion variations and not to regional variations in the warning system. In the 1980s, reporting of locusts attacks was mainly based on phone calls of people that observed locust swarms in the place they live. It is possible that in some areas observations are less exhaustive than in other places, or that people declare only the name of the village they live in. It could also be the case that people reporting were better informed than others about the existence of the Desert Locust Information Service or were expecting help from the government following the attack.
Table 2 shows the average
population size of urban and rural localities according to their treatment status and for the cohort 1988. The fact that we observe that the locusts aected localities are, on average and in 1998, more urban than others conrms the previous hypothesis. This could create a self reporting bias, which sign is not obvious. It depends on the proportion of urban localities in the whole sample and if the shock has been more or less intensive in urban versus rural localities.
For this reason, we run regressions separately for rural and urban localities.
In
any case, incomplete observation of swarms attacks will lead to an under-evaluation of the impact, as some of the villages taken as controls will also be aected by the locust plague.
[insert table 2 and gure 3 about here]
4
Thanks to the geo-referencing of each locality,
we match its coordinates with rainfall
data from the Climate Research Unit (CRU) at the University of East Anglia. Precipitation levels are available from 1901 to 2006 on a month-by-month basis with a precision of 0.5x0.5 degree.
We compute annual rainfall shocks for each locality, as the dierence between the
natural log of precipitation at time
t
and the natural log of mean annual precipitations
calculated over the 1940-1998 period. Given that rainfalls are likely to aect the welfare of households, particularly in the rural areas, and to control for the potential correlation between locust invasion and high precipitations we compute the rainfall shock variables ten years in a row, starting three years before the birth date and ending seven years after the birth date
4 Actually,
the 1998 census data does not provide the coordinates of 1,200 localities (among 10,000) mostly located in northern Mali. We complete the coordinates of the dataset only for localities aected by locust swarms.
14
5
when individuals are in age of school admission.
We implement this specication for school
enrollment. When dealing with grade attainment or primary level achievement, we complete the model with rainfall shock variables occurring between age 8 and 13 and that may inuence the educational attainment of shocked individuals.
4
Results
The main results are presented in tables 3 to 5. To assess, as far as possible, the heterogeneity of the impact, we rst distinguish between boys and girls educational outcomes and then we discriminate between rural and urban localities. In order to save space, we chose to report only the coecients of the cohort times locust invasion dummy variable.
All regressions
include controls for rainfalls, together with birth cohort dummies and village xed eects. Robust standard errors are reported. Table 3 shows the results obtained when average school enrollment at the locality level is the dependent variable. In table 4 the dependent variable is the average grade attainment and, in table 5, the proportion of children that completed primary school both among those enrolled. For all three tables, columns 1 to 3 (resp. 4 to 6) show the results obtained rst for the full population of boys (resp. girls) in column 1 (resp. 4), then separately for urban and rural areas (columns 2 and 3 for boys; 5 and 6 for girls).
Looking rst at table 3 for boys and girls in Mali as a whole (columns 1 and 4), the striking result is the strong and signicant negative impact of locust swarms on the enrollment of children born after 1982. The strongest impact is found for cohorts 1988 and 1989, that is
6
for children that were potentially in-utero and up to two years old during the locusts invasion . For boys, the proportion of children born in 1988-1989 ever enrolled at school is reduced by 6.1 percentage points if they lived in a community invaded by locusts. For girls the impact size is smaller: 3.5% but remains signicant. In relative terms the impact on each gender is of similar amplitude, with a 25% decrease in the proportion of enrolled children from cohort 1989.
As locusts eat the harvests of farmers and the food of cattle one expects their impact to be higher in rural than in urban areas. This is what we nd, as can be seen for boys and
5 Since
children enter school at seven years old, we control for up to seven years after the birth date, in order to account for any impact that rainfall variations might have on school enrollment. 6 The equality hypothesis between coecients of cohorts 1990-88 and 1983-85 is rejected which corroborates the fact that the locusts plague had a heterogeneous impact on the enrollment of children, diminishing with age. These results are observed for boys and for girls at the full sample level, as well as at more disaggregated ones (tests not shown). 15
girls in columns 3 and 6 of table 3.
No eect is found in urban areas (columns 2 and 5),
which conrms that the partial destruction of harvests had no sizable macroeconomic eect. In rural areas on the contrary, the eect is found stronger than in Mali at large. The decrease in the proportion of enrolled children rises from 6 to 7.5 percentage points and from 3.5 to 5, respectively for boys and girls born in 1989.
Also striking is the fact that before 1983 for boys and for girls living in rural areas, the cohort times locust invasion interaction dummy coecient is never found signicant on school enrollment. In Mali school normally starts at 7. Children born in 1983 were at most 6 in 1988 and 7 in 1989, so it is not obvious to explain why their school enrollment should be lower than that of children born one year earlier. However, as we have seen, people are relatively imprecise when reporting their age and we observe peaks in the age distribution around multiples of 5. People born in 1983 were 15 in 1998. Because of reporting mistakes, many of those that declared being 15 in 1998 were in fact born earlier than 1983.
This
could explain why the 1982 cohort coecient is not found negative if locusts invasions have a negative eect on the probability to enter school and if those children that did not enroll are also more likely to report their age less precisely. In order to check for this explanation gure 6 reports the average cohort size at the locality level for enrolled and non enrolled children separately. If our intuition is correct, then one should observe more pronounced peaks around cohorts that, in 1998, correspond to an age that is a multiple of ve (that is 1973, 1978, 1983, 1988) in the uneducated population than in the educated one. The results are striking and conrm our intuition: the curve for the enrolled population appears much smoother than that of the unenrolled population and the dierence is larger precisely for the birth years that are 25, 20, 15 and 10 years before 1998. Such reporting mistakes could also explain why those that were declared born in 1990 and 1991 are also found negatively impacted, though the swarms attack occurred after their reported birth date. The other possibility being a strong and negative impact on those children that were in-utero when the invasion happened.
Table 4 presents the results on grade attainment. While the impact appears weak in the whole population, we nd that for all girls cohorts born after 1977 (column 6), exception made of cohort 1987, the number of completed years of schooling is lower if in 1988-1989 they lived in a rural community attacked by locusts. The major signicant eect at the 1% level is found for cohort 1981 which completed up to one lower grade than the reference cohort
7
(1969) . Grade attainment of boys that were in age of entering school in rural areas at the
7 Ceteris
paribus
16
time of plague is impacted. The amplitude of the impact is lower than for girls, however grade attainment of cohort 1981 is reduced by 0.43 grade for boys in invaded localities. Looking now at the coecients obtained when the dependent variable is the proportion of enrolled children that completed primary school in the locality (table 5) we nd that a negative impact is found for boys and girls in age of entering school at the time of the plague i.e cohorts 1980-1982: for cohort 1981 in rural areas the proportion of boys and girls that completed primary school among those enrolled is reduced by 16 and 13%, when compared with the reference cohort (columns 3 and 6).
What could explain these results ? The channels through which locust swarms could impact school enrollment are twofold. First, impoverished, stricken households could decide to keep their child at home in response to a need for labor.
If this is the case, then we
should observe that older children have a lower educational attainment, as they are also likely to be withdrawn from school.
Second, the lack of food that could follow from the locusts
invasion might have a negative and durable impact on the strength and cognitive abilities of children.
In face of this, household might decide not to enroll them.
For rural girls,
the results for educational attainment and primary level achievement (columns 6, tables 4 and 5), validates the rst line of explanation, without excluding the second. Enrolled boys educational attainment, on the other hand, does not seem much impacted (columns 3, tables 4 and 5), except for cohorts 1980 and 1981. Their primary school achievement is also largely aected, similarly to girls'.
These results conrm that school admission time is a crucial
period in educative life. Educational outcomes of children experiencing the shock while they were in age to enter school have been more impacted than others.
Findings also conrm
that, especially in the rural areas, girls schooling achievement seems to be more sensitive to household's resources diminution than boys'.
5
Robustness checks
To strengthen the condence in our results, we perform in this section some robustness checks. We rst address the resident selection issue in our population and provide answers on how this aects our results through more detailed descriptive statistics and the implementation of a strategy to hold account for migration. Then, we check the robustness of our results to variations in the cut-o point cohort sample. Finally, we test whether not taking account of rainfalls aects the estimated impacts.
17
5.1 Resident bias Migrants may be a peculiar category of individuals within the population and their decision to migrate might be correlated with these specic characteristics and/or with locust invasion. Hence, a more precise discussion on their characteristics and their potential divergence among treated and untreated localities can be helpful to understand our ndings. Migrants are dened as people whose age in 1998 is higher than the duration of residence in their present place of living. As every individual was asked about its birth place and, if
cercle, we test the robustness of our results, by reallocating migrants cercle and, inside the cercle, by choosing randomly their locality of birth.8 A
in Mali, about its birth in their birth
total of 40 simulations are done. Table A.1 sets migrants characteristics according to their gender, living area at birth time, group of birth cohorts
9
and treatment status. The number
of migrants per locality and the migrants school enrollment are averages computed over the 40 simulations. For school enrollment we test for the dierence in the average enrollment of migrants between locust aected and non aected localities (test 1) and between migrants and non migrants in locust aected and non aected localities (test 2). For each test we report the number of times the dierence is found statistically signicant over the 40 simulations. As the proportion of migrants is small relative to that of non migrants, we take the estimated enrollment rates for the non migrant population as the reference to which the migrants enrollment rate should be compared.
Unsurprisingly the results show that in rural
areas migrants are much more educated than non migrants, both in locust invaded and non invaded localities. In urban areas no such dierence is found. If we now compare the estimated enrollment rates between locust invaded and non invaded localities, the non migrant enrollment rate show that in rural areas, locust invaded localities exhibit signicantly lower rates, consistently with what is shown in the top panel of gures 4 and 5. However this is not the case for migrants: those coming from locusts invaded localities have a higher or an equivalent rate of school enrollment than those coming from non invaded localities. Thus migration appears to have been selective, indeed, with the non migrant population more likely to be less educated than the population at large in invaded localities. For school enrollment, this selective migration is likely to downward bias our estimated impact of the locust plague. For over educational outcome, the prediction is less clear: either migrants are impacted at least as much as non migrants which potentially motivates their decision to move, then we would
8 For
each locality the probability to be selected among all localities of a given cercle is based on its relative population. Reallocation of migrants among cercles depends on each cercle emigration rate. 9 We distinguish between two groups of cohorts : 1980-1991 and 1979-1965, in order to identify potential heterogeneity that might be linked to dierent educational or economic environments. 18
underestimate the impact, or migrants, being a more educated hence reactive population, able to leave and adapt some place else, are less impacted than non migrants, then we would overestimate the impact. In order to assess the amplitude and direction of the possible biases, for each of the 40 simulated reallocations of migrants, we estimate our model on the resulting simulated population and check whether it signicantly changes our results.
Figure 7 for boys and
gure 8 for girls show, for each cohort, the 95% condence interval of the locusts estimated impact on school enrollment when migrants are reallocated within their birth
cercle, together
with the median of the 40 estimates, when it is signicantly dierent from zero at least one time.
Dierent markers are employed depending on the proportion of non zero estimates.
When signicant we also add to this graph the estimated coecients found with the non migrant population and reported in table 3.
What can be seen at rst is that with the exception of cohorts 1984 for girls and 1991 for boys, the coecients estimated on the non migrant population always lies within the bounds of the 95% condence interval built from the estimations obtained from the simulated populations. The median of the simulated coecients is also found very close to the estimates and for cohorts 1983 to 1990 for boys and 1986, 1988, 1989 and 1990 for girls, the simulated coecients are found signicantly dierent from zero in at least 90% of the cases in the total and rural populations. For the urban population, the simulated results are also coherent with our estimates, since the proportion of non zero simulated coecients is only higher than 50% for cohort 1983 and for the boys sample, for which the estimate is also found signicant. For girls, the simulated coecients are never found signicant for the urban population, which conrms our estimates and the results are not reported in the graphs. The only signicant discrepancy is for cohort 1968 in the boys population, for which a signicant coecient is found in more than 90% of the simulations, while the estimate reported in table 3 is not signicant. Similar results are obtained for other educational outcomes (results not shown). We can then conclude that holding account for selective migration does not alter our previous results :
school enrollment of children potentially in-utero or in early childhood
during the shock is the most impacted while educational attainments are lowered for children in age to enter school at the time of the locust plague. Results are driven by rural areas and suggest that the shock did not have lasting consequences on the education of people from urban areas.
19
5.2 Cut-o point cohort sample We further check the robustness of our identication strategy by testing whether the observed results would be driven by the arbitrary cut-o point cohort (cohort 1965) of the sample. To perform our Dierence-in-Dierence strategy correctly, we rst need to identify non impacted individuals within treated localities and compare their education with that of potentially aected ones. Non impacted individuals within aected localities are individuals that were "too old" during the shock for their education to be impacted by the plague.
Hence we
consider that the education of children aged more than eighteen during the shock, i.e cohorts born before 1971, could not have been aected. We decide to include in our sample cohorts up to 1965, since the education of older cohorts may have been impacted by the previous locust plague which ended in 1962. Doing so we also limit dierences in the environmental contexts between potentially aected and non aected cohorts. However, we check whether this decision inuences our results. We run our specications on 4 populations, dierent from our base one and allow the cut-o point cohort to vary from 1966 to 1969.
Findings attest that our results are robust to variation of the cut-o point
cohort (tables A.2 to A.5). Between cohorts and outcomes, the same pattern is found for all specications.
5.3 Rainfalls matter As there might be a correlation between the occurrence of locusts invasions and rainfalls, and since rainfalls have been shown to have a non zero impact on health and education in previous studies (eg. Maccini and Yang 2009), our baseline specication controls for rainfall variations around the time of birth. When we look at treatment eect coecients, both with and without controlling for rainfalls shocks around the time of birth (table A.6), we notice that coecients are larger without controlling for rainfalls. On average, coecients are overestimated by one percentage point for boys and 0.5 for girls. This goes in the sense of a global negative impact of rainfalls on children education in Mali, exception made of Bamako. Controlling for rainfalls is therefore a necessary condition when dealing with long run impacts of economic shocks in developing countries. In light of previous results and especially with regard to the amplitude of rainfall shocks eects on education, we decide to test the robustness of our results by adding rainfalls shocks that occurred at the time of invasion (1987-1989) for all cohorts in our specication. These shocks happened at a dierent time in life for each cohort. We nd that adding control for rainfalls shocks during the locusts plague does not alter our results, treatment eect coecients are slightly stronger, still in a negative
20
way (results not shown).
6
Conclusion
This paper nds that the large and negative income shock induced by the 1987-1989 locust plague in Mali has long run impact on educational enrollment and completion of children who experienced the shock at a critical time of their childhood.
The identication strategy is dened at the village level and assimilates the shock as a "treatment". Therefore, we propose a dierence in dierence within village strategy which allows us to identify the impact of the locust plague on average educational outcomes per village, exploiting the geographical variation of locust invasions. In our study, we allow for a heterogeneous impact of shocks along age and sex and pay particular attention to dierences between urban and rural households.
We nd a clear and strong impact on school enrollment of children born or aged less than seven years old at the time of shock.
Children born in 1988-1989, the main years of
invasion, are those whose school enrollment has been the most aected by the plague. Boys are more strongly aected than girls, but on the other hand, girls schooling achievement seems to be more sensitive to the shock as we nd a signicant and negative impact on the grade achieved for all cohorts born after 1977. We can attribute this mitigated impact to the fact that boys' education is considered more of a priority than girls'. As we expected the impact in rural areas is much stronger and signicant than that in urban areas, which conrms the low macroeconomic impact of locust invasions. A negative impact is also clearly detected on educational attainment of children that were in age to enter school during the plague. Indeed, treatment eect on grade attainment is broadly twice for girls cohort 1981 than for cohorts born two years before or after.
Our results reveal a strong impact of economic shocks on the education of children impacted, especially those experiencing it during their earliest childhood. They also suggest that at least part of the adjustment seems to have happened at the nutritional level, impacting on the long run children who were at an early stage of development and girls who are more vulnerable members within a household.
Although enrollment of children in age to enter
school at the time of shock does not seem impacted, educational attainment of both gender has been deteriorated, mainly in rural areas.
The dierence in impacts between boys and
girls claims that some consequences result from a discriminative behavior. Also some impacts
21
could have been lowered if insurance scheme had been provided to vulnerable households. This paper contributes to the literature by showing that consumption smoothing is not completely possible even when facing an idiosyncratic shock, as this is the case for the 1987-1989 locust plague (Skouas
et al., 1997).
Further work, using data on health and nutrition status, will help to precise these results, as it will allow to better identify the channels through which locust invasions impact education.
22
References [1] Most of
the references concerning
Desert
Locusts can
be
found at:
http://ispi-
lit.cirad.fr/index/LocustLit.htm
No Small Matter: The Impact of Poverty, Shocks, and Human Capital Investments in Early Childhood Development. Human development perspectives, The
[2] Alderman, H. 2011.
World Bank, Washington DC, 208 p.
[3] Alderman, H., Hoddinott J. and Kinsey B., 2006. "Long term consequences of early childhood malnutrition."
Oxford Economic Papers
58, 450-473.
[4] Almond D., 2006. "Is the 1918 Inuenza Pandemic Over? Long-Term Eects of In Utero Inuenza Exposure in the Post-1940 U.S. Population."
Journal of Political Economy, vol.
114(4), pp. 672-712.
[5] Banerjee, A., Duo E., Postel-Vinay G., Watts T., 2007. Long Run health Impacts of Income Shocks: Wine and Phylloxera in 19th Century France. NBER Working Paper 12895.
British
[6] Barker, D.J.P. 1992. "Fetal and Infant origins of Later Life Disease" London,
medical journal. [7] Case A., Paxson C.,2010. "Causes and Consequences of Early Life Health," NBER Working Paper No. 15637.
[8] Case A., Fertig A., Paxson C., 2005. "The Lasting Impact of Childhood Health and Circumstance."
Journal of Health Economics
24 (2), 265-289.
[9] Currie J., Moretti E., 2007. "Biology as Destiny? Short- and Long-Run Determinants of Intergenerational Transmission of Birth Weight."
Journal of Labor Economics
25 (2),
231-263.
[10] Currie, J. 2009. "Healthy, Wealthy and Wise: Childhood, and Human Capital Development."
Socioeconomic Status, Poor Health in
Journal of Economic Literature
47(1):
87-122.
[11] Dercon S. 2004. "Growth and Shocks: evidence from Rural Ethiopia", (2004),
of Development Economics, August, vol 74 (2), pp. 309-29.
23
Journal
[12] Diarra, S.O., Diakite Y., Konate M. K. and Lange M-F. 2001. Politiques éducatives et
La demande d'éducation en Afrique: état des connaissances et perspectives de recherche, the Union système éducatif actuel au Mali. chap. 8 in M. Pilon and Y. Yaro (ed.)
for African Population Studies, publication n 1.
[13] Duranton J-F. and Lecoq M., 1990.
Le criquet pélerin au Sahel.
Collection Acridologie
Operationelle No. 6, CIRAD-PRIFAS, Montpellier, pp 183.
[14] Ferreira F. H. G. and Schady N., 2009." Aggregate Economic Shocks, Child Schooling and Child Health.",
The World Bank Research Observer, Vol 24(2), pp. 147-181.
[15] Gorgens, T., Meng X. and Vaithianathan R. , 2011. "Stunting and selection eects of famine: a case study of the Great Chinese Famine."
Journal of Development Economics,
Article in press.
[16] Grimard, F. and Laszlo S., 2010. Long term eects of civil conict on women's health outcomes in Peru. mimeo.
[17] Herok C.A., Krall S. 1995.
Economics of Desert Locust Control. GTZ, Eschborn, 66 pp.
[18] Jacoby H. and Skouas E. 1997. "Risk, nancial markets and human capital in a developing country",
[19] Joe, S. 1997.
Review of Economic Studies, vol.64, pp. 311-335
Economic and policy issues in Desert Locust management: a preliminary
analysis, FAO/EMPRES Workshop on Economics in Desert Locust Management, Cairo, September
[20] Lange M.-F. and Diarra S. O., 1999. "Ecole et démocratie : 1'"explosion" scolaire sous la IIIe République au Mali", Politique africaine, n'76, décembre, pp.164-172.
[21] Latchininsky A.V., Launois-Luong M.H. 1997. Le Criquet pélerin (
Schistocerca gregaria
Forskal, 1775) dans la partie nord-orientale de son aire d'invasion. CIRAD-PRIFAS, Montpellier, France.
[22] Lecoq, M. 2004. Vers une solution durable au problème du criquet pélerin ?,
Sècheresse,
15 (3) : 217-24.
Civil Conict and Human Capital Accumulation: The Long Term Eects of Political Violence in Peru. BREAD Working paper nï¾½ 24, September.
[23] Leon, G. 2009.
24
[24] Maccini, S. Yang, D. 2009. "Under the Weather: Health, Schooling, and Economic Consequences of Early-Life Rainfall,"
American Economic Review,
vol. 99(3), pp. 1006-26,
June.
[25] Soumare A. M. 1994. " Factors that Aect Girls' Access of and Retention in School in Mali", Projet, Washington, Academy for Educational Development.
[26] Thomson, A. and Miers H. 2002. Assessment of the socio-economic impact of Desert Locusts and their control, nal report, Oxford Policy Management, April.
[27] University
of
East
Anglia
Climate
Research
[Internet]. British Atmospheric Data Centre, http://badc.nerc.ac.uk/data/cru
25
Unit 2008,
(CRU).
2008.
CRU
Datasets,
Date of citation. Available at
Figure 1: Locust invasion in Africa
Source: http://www.cnlcp.net/
26
Figure 2: Crop and food production indexes
Source:http://countrystat.org/mli/cont/pxwebquery/ma/133cpd010/fr, authors' calculations.
27
28
Reading: Each dot corresponds to a Malian locality listed by the FAO's Desert Locust Information Service (DLIS) as aected by locust swarms between 1987 and 1989.
Figure 3: Malian Localities aected by the 1987-1989 locust plagues
Figure 4: Educational variables, Mali, Boys born in 1965 - 1991
Note: These graphs are computed on a sample of people that never moved from the place they live in 1998. Moreover, people that live in Bamako are excluded from the sample. COHORT identies the birth of year i.e COHORT 1981 identies individuals born in 1981, allowed to enter school from 1988 (7 years old) and aged 17 in 1998, year of data collection used for our calculations. Source: Malian Population Census data, 1998, our own calculation.
29
Figure 5: Educational variables, Mali, Girls born in 1965 - 1991
Note: These graphs are computed on a sample of people that never moved from the place they live in 1998. Moreover, people that live in Bamako are excluded from the sample. COHORT identies the birth of year i.e COHORT 1981 identies individuals born in 1981, allowed to enter school from 1988 (7 years old) and aged 17 in 1998, year of data collection used for our calculations. Source: Malian Population Census data, 1998, our own calculation.
30
Figure 6: Coecients comparison - Samples with and without migrants
Note: Simulations randomly assign migrants in a locality belonging to their birth cercle, weighted by its relative population within cercle. Simulations are performed 40 times. Cohort reference : 1969. Source: Malian Population Census data, 1998, our own calculation.
31
Figure 7: Coecients comparison - Samples with and without migrants
Note: Simulations randomly assign migrants in a locality belonging to their birth cercle, weighted by its relative population within cercle. Simulations are performed 40 times. Cohort reference : 1969. Graph for urban girls not shown due to lack of signicant coecients. Source: Malian Population Census data, 1998, our own calculation.
32
Table 1: Number of treated and controlled localities and average number of individuals by cohort. Cohort
1991 1990 1989 1988 1987 1986 1985 1984 1983 1982 1981 1980 1979 1978 1977
Locust localities
Other localities
/Treatment
/Control
Total
group
group
954
9 048
(37)
(28)
(29)
957
9 040
9 997
10 003
(37)
(27)
(28)
932
8 948
9 880
(25)
(20)
(20)
960
9 024
9 984
(35)
(24)
(25)
879
8 827
9 706
(21)
(16)
(16)
938
8 942
9 880
(31)
(22)
(23)
899
8 903
9 802
(23)
(17)
(18)
908
8 879
9 787
(24)
(17)
(18)
943
8 989
9 932
(29)
(20)
(21)
901
8 852
9 757
(24)
(16)
(16)
905
8 786
9 687
(20)
(14)
(14)
942
8 930
9 872
(27)
(18)
(19)
826
8 459
9 285
(27)
(18)
(10)
952
8 965
9 917
(32)
(19)
(20)
789
8 262
9 917
(12)
(9)
(9)
Notes: Average number of individuals per cohort are in brackets (boys and girls aggregated).
33
Table 1 continued Cohort
1976 1975 1974 1973 1972 1971 1970
Locust localities
Other localities
/Treatment
/Control
Total
group
group
909
8 733
(17)
(12)
(13)
826
8 406
9 232
9 642
(12)
(9)
(9)
815
8 210
9 025
(11)
(8)
(9)
944
8 838
9 782
(26)
(15)
(16)
839
8 391
9 230
(12)
(8)
(9)
855
8 411
9 266
(12)
(8)
(9)
891
8 628
9 519
(16)
(10)
(11)
1969
742
7 650
8 392
(8)
(6)
(6)
1968
951
8 934
9 885
(31)
(16)
(18)
1967
681
7 525
8 206
(7)
(6)
(6)
1966
876
8 410
9 286
1965
(12)
(8)
(9)
751
7 847
8 598
(9)
(6)
(7)
Notes: Average number of individuals per cohort are in brackets (boys and girls aggregated).
34
Table 2: Breakdown of the sample according to urban and rural areas(a) . Locust localities
Other localities
/Treatment
/Control
group Urban localities Rural localities Total
Total
group
74
263
337
(168)
(113)
(125) 9 647
886
8 761
(24)
(22)
(22)
960
9 024
9984
(35)
(25)
(26)
Notes: Average number of individuals per locality are in brackets (boys and girls aggregated). (a): Cohort 1988.
35
Table 3: Impact of locust invasion on school enrollment, boys and girls. (1) (2) (3) (4) Full sample Urban Rural Full sample Boys Boys Boys Girls Born in locust loc. year 91 -0.0176* -0.0120 -0.0257** -0.0184***
(5) Urban Girls 0.00537
(6) Rural Girls -0.0267***
(0.0102)
(0.0300)
(0.0108)
(0.00640)
(0.0231)
(0.00660)
Born in locust loc. year 90
-0.0542***
-0.0480
-0.0639***
-0.0349***
0.0163
-0.0482***
(0.0102)
(0.0310)
(0.0108)
(0.00712)
(0.0269)
(0.00715)
Born in locust loc. year 89
-0.0614***
-0.0174
-0.0752***
-0.0339***
0.0333
-0.0504***
(0.0108)
(0.0330)
(0.0112)
(0.00784)
(0.0269)
(0.00787)
Born in locust loc. year 88
-0.0602***
-0.0189
-0.0731***
-0.0351***
-0.0116
-0.0464***
(0.0105)
(0.0297)
(0.0110)
(0.00701)
(0.0252)
(0.00713)
Born in locust loc. year 87
-0.0537***
-0.0180
-0.0705***
-0.0256***
0.0356
-0.0445***
(0.0114)
(0.0332)
(0.0118)
(0.00832)
(0.0274)
(0.00835)
Born in locust loc. year 86
-0.0456***
-0.00171
-0.0600***
-0.0243***
0.00197
-0.0359***
(0.0110)
(0.0317)
(0.0115)
(0.00719)
(0.0265)
(0.00731)
Born in locust loc. year 85
-0.0433***
-0.00397
-0.0593***
-0.00866
0.0279
-0.0213***
(0.0105)
(0.0333)
(0.0108)
(0.00743)
(0.0248)
(0.00761)
-0.0316***
-0.0158
-0.0453***
-0.0124*
0.0146
-0.0233***
(0.0111)
(0.0346)
(0.0115)
(0.00678)
(0.0246)
(0.00694)
-0.0324***
-0.0604**
-0.0411***
-0.00942
-0.00653
-0.0164***
(0.00991)
(0.0294)
(0.0104)
(0.00594)
(0.0218)
(0.00615)
Born in locust loc. year 84 Born in locust loc. year 83 Born in locust loc. year 82 Born in locust loc. year 81 Born in locust loc. year 80 Born in locust loc. year 79 Born in locust loc. year 78 Born in locust loc. year 77
0.00999
-0.0105
0.00254
0.00669
0.00762
0.00299
(0.00986)
(0.0323)
(0.0103)
(0.00569)
(0.0226)
(0.00586)
-0.000288
-0.0336
-0.00754
0.00318
0.0213
-0.00195
(0.00950)
(0.0328)
(0.00987)
(0.00581)
(0.0205)
(0.00606)
0.00872
0.0130
0.000101
0.00883
0.00510
0.00578
(0.00985)
(0.0309)
(0.0104)
(0.00567)
(0.0212)
(0.00588)
0.00740
0.00153
-0.00239
0.00940
0.0276
0.00254
(0.0105)
(0.0310)
(0.0112)
(0.00635)
(0.0271)
(0.00639)
-0.00320
-0.0242
-0.00756
-0.000416
-0.00794
-0.00157
(0.00940)
(0.0297)
(0.00997)
(0.00531)
(0.0212)
(0.00551)
0.0145
0.0231
0.00616
0.000197
0.0140
-0.00493
(0.0112)
(0.0341)
(0.0118)
(0.00651)
(0.0232)
(0.00684)
36
Table 3 continued. (1) Full sample Boys -0.000560
(2) Urban Boys -0.0109
(3) Rural Boys -0.00526
(4) Full sample Girls 0.00544
(5) Urban Girls 0.00142
(6) Rural Girls 0.00446
(0.0101)
(0.0336)
(0.0107)
(0.00565)
(0.0205)
(0.00595)
Born in locust loc. year 75
0.000717
0.00398
-0.00383
-0.00306
0.00270
-0.00584
(0.0110)
(0.0325)
(0.0117)
(0.00586)
(0.0209)
(0.00614)
Born in locust loc. year 74
-0.00725
-0.00285
-0.0100
0.00356
0.0175
0.000733
(0.0105)
(0.0291)
(0.0113)
(0.00663)
(0.0249)
(0.00687)
0.000438
-0.0235
-8.86e-05
-0.00252
0.00516
-0.00220
Born in locust loc. year 76
Born in locust loc. year 73
(0.00998)
(0.0258)
(0.0107)
(0.00535)
(0.0201)
(0.00563)
0.00508
-0.00136
0.00387
-0.00463
0.0159
-0.00722
(0.0109)
(0.0304)
(0.0116)
(0.00590)
(0.0204)
(0.00622)
Born in locust loc. year 71
-0.0109
-0.000529
-0.0128
-0.00481
-0.0110
-0.00380
(0.0107)
(0.0322)
(0.0114)
(0.00608)
(0.0231)
(0.00631)
Born in locust loc. year 70
-0.0119
-0.0431
-0.0100
-0.00609
-0.00572
-0.00562
(0.0103)
(0.0307)
(0.0109)
(0.00561)
(0.0205)
(0.00588)
Born in locust loc. year 72
Born in locust loc. year 68 Born in locust loc. year 67 Born in locust loc. year 66 Born in locust loc. year 65 Constant Rainfall control variables Fixed eect locality Fixed eect cohort Observations Number of localities R2
-0.0128
-0.00449
-0.0138
0.000534
0.00722
0.00140
(0.00981)
(0.0278)
(0.0105)
(0.00558)
(0.0222)
(0.00581)
-0.00183
0.0393
-0.00659
0.0125
0.0203
0.0134
(0.0127)
(0.0342)
(0.0137)
(0.00819)
(0.0229)
(0.00889)
1.11e-05
0.00421
-0.000622
-0.000200
-0.00568
0.00175
(0.0105)
(0.0323)
(0.0112)
(0.00630)
(0.0221)
(0.00665)
-0.00242
0.0452
-0.00945
0.0107
0.0207
0.0122
(0.0117)
(0.0342)
(0.0125)
(0.00842)
(0.0240)
(0.00906)
0.129***
0.325***
0.122***
0.0498***
0.235***
0.0429***
(0.00303)
(0.0111)
(0.00313)
(0.00190)
(0.0101)
(0.00193)
yes yes yes 229 369 10 111 0.087
yes yes yes 8 685 340 0.322
yes yes yes 220 684 9 771 0.082
yes yes yes 236 733 10 112 0.097
yes yes yes 8 770 340 0.414
yes yes yes 227 963 9 772 0.090
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
37
Table 4: Impact of locust invasion on grade attainment, boys and girls. (1) (2) (3) (4) Full sample Urban Rural Full sample Boys Boys Boys Girls Born in locust loc. year 91 -0.114 0.0317 -0.00551 -0.365***
(5) Urban Girls 0.430**
(6) Rural Girls -0.524***
(0.125)
(0.167)
(0.171)
(0.139)
(0.206)
(0.192)
Born in locust loc. year 90
-0.0840
0.0260
-0.00306
-0.262*
0.395**
-0.415**
(0.123)
(0.164)
(0.170)
(0.138)
(0.197)
(0.192)
Born in locust loc. year 89
-0.0850
-0.00790
-0.0273
-0.188
0.372*
-0.339*
(0.124)
(0.169)
(0.170)
(0.136)
(0.191)
(0.190)
Born in locust loc. year 88
0.00288
-0.0824
0.0610
-0.228*
0.282
-0.431**
(0.123)
(0.163)
(0.170)
(0.135)
(0.200)
(0.185)
Born in locust loc. year 87
-0.0560
-0.0932
-0.0262
-0.131
0.225
-0.320
(0.124)
(0.168)
(0.170)
(0.144)
(0.203)
(0.198)
Born in locust loc. year 86
-0.0361
-0.189
-0.00735
-0.199
0.172
-0.418**
(0.122)
(0.158)
(0.169)
(0.137)
(0.192)
(0.189)
Born in locust loc. year 85
-0.119
-0.190
-0.137
-0.260*
0.343*
-0.572***
(0.126)
(0.168)
(0.175)
(0.143)
(0.193)
(0.198)
-0.139
-0.327*
-0.106
-0.352**
0.138
-0.631***
(0.129)
(0.178)
(0.176)
(0.152)
(0.231)
(0.200)
-0.229*
-0.256
-0.257
-0.260*
0.258
-0.582***
(0.127)
(0.167)
(0.173)
(0.147)
(0.209)
(0.195)
-0.104
-0.107
-0.203
-0.379**
0.225
-0.801***
(0.146)
(0.176)
(0.202)
(0.166)
(0.208)
(0.225)
-0.267*
-0.168
-0.433**
-0.493***
0.380*
-1.036***
(0.142)
(0.172)
(0.198)
(0.166)
(0.211)
(0.222)
-0.264*
-0.197
-0.351*
-0.315**
0.206
-0.615***
(0.138)
(0.176)
(0.189)
(0.154)
(0.197)
(0.211)
-0.0953
-0.0305
-0.233
-0.215
0.126
-0.471**
(0.138)
(0.166)
(0.195)
(0.148)
(0.179)
(0.208)
-0.0747
-0.234
-0.0696
-0.191
0.252
-0.454**
(0.130)
(0.176)
(0.177)
(0.156)
(0.181)
(0.221)
-0.00541
-0.184
0.0274
-0.0117
0.379
-0.294
(0.143)
(0.172)
(0.202)
(0.181)
(0.233)
(0.256)
Born in locust loc. year 84 Born in locust loc. year 83 Born in locust loc. year 82 Born in locust loc. year 81 Born in locust loc. year 80 Born in locust loc. year 79 Born in locust loc. year 78 Born in locust loc. year 77
38
Table 4 continued. (1) Full sample Boys 0.0556
(2) Urban Boys -0.127
(3) Rural Boys 0.0850
(4) Full sample Girls -0.146
(5) Urban Girls 0.191
(6) Rural Girls -0.371
(0.143)
(0.193)
(0.198)
(0.158)
(0.186)
(0.226)
Born in locust loc. year 75
-0.0274
-0.152
0.0179
-0.209
0.221
-0.420*
(0.150)
(0.209)
(0.206)
(0.172)
(0.220)
(0.244)
Born in locust loc. year 74
-0.0312
-0.209
0.0119
-0.0679
0.262
-0.218
(0.145)
(0.175)
(0.205)
(0.170)
(0.246)
(0.237)
0.0293
-0.0617
0.0376
-0.199
0.264
-0.431**
(0.134)
(0.151)
(0.188)
(0.157)
(0.212)
(0.213)
0.203
0.0453
0.283
-0.227
0.164
-0.482**
(0.144)
(0.188)
(0.203)
(0.172)
(0.236)
(0.243)
Born in locust loc. year 71
0.0703
-0.174
0.166
-0.128
0.258
-0.332
(0.142)
(0.185)
(0.197)
(0.167)
(0.227)
(0.235)
Born in locust loc. year 70
-0.0156
-0.372**
0.102
-0.143
0.283
-0.386
(0.138)
(0.188)
(0.187)
(0.172)
(0.231)
(0.243)
-0.0890
0.102
-0.149
-0.204
0.114
-0.360*
(0.139)
(0.183)
(0.188)
(0.151)
(0.200)
(0.207)
-0.0900
0.0546
-0.145
-0.151
0.309
-0.455*
(0.151)
(0.210)
(0.211)
(0.174)
(0.231)
(0.242)
0.0472
-0.153
0.137
0.174
0.105
0.133
(0.143)
(0.184)
(0.195)
(0.161)
(0.188)
(0.238)
-0.187
-0.161
-0.173
-0.0249
0.251
-0.176
(0.158)
(0.181)
(0.229)
(0.172)
(0.208)
(0.257)
4.413***
4.969***
4.365***
4.033***
4.849***
3.910***
(0.0416)
(0.0805)
(0.0461)
(0.0528)
(0.0866)
(0.0621)
yes yes yes 81 444 7 797 0.451
yes yes yes 6 904 317 0.719
yes yes yes 74 540 7 480 0.434
yes yes yes 56 647 6 966 0.386
yes yes yes 6 548 314 0.637
yes yes yes 50 099 6 652 0.363
Born in locust loc. year 76
Born in locust loc. year 73 Born in locust loc. year 72
Born in locust loc. year 68 Born in locust loc. year 67 Born in locust loc. year 66 Born in locust loc. year 65 Constant Rainfall control variables Fixed eect locality Fixed eect cohort Observations Number of localities R2 *** p<0.01, ** p<0.05, * p<0.1
Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
39
Table 5: Impact of locust invasion on primary level achievement, boys and girls. (1) (2) (3) (4) (5) Full sample Urban Rural Full sample Urban Boys Boys Boys Girls Girls Born in locust loc. year 85 -0.0160 -0.0353 -0.0228 0.0116 0.0682**
(6) Rural Girls -0.0119
(0.0184)
(0.0297)
(0.0228)
(0.0214)
(0.0313)
(0.0282)
Born in locust loc. year 84
-0.0158
-0.0740**
-0.0214
-0.0246
0.0198
-0.0613**
(0.0214)
(0.0322)
(0.0264)
(0.0240)
(0.0353)
(0.0309)
Born in locust loc. year 83
-0.0190
-0.0569
-0.0366
-0.00113
0.0494
-0.0517*
(0.0218)
(0.0351)
(0.0263)
(0.0244)
(0.0369)
(0.0303)
Born in locust loc. year 82
-0.00848
-0.0486
-0.0521
-0.0193
0.000368
-0.0804**
(0.0289)
(0.0432)
(0.0357)
(0.0315)
(0.0448)
(0.0401)
Born in locust loc. year 81
-0.0876***
-0.0561
-0.160***
-0.0399
0.0431
-0.132***
(0.0308)
(0.0426)
(0.0381)
(0.0323)
(0.0457)
(0.0404)
Born in locust loc. year 80
-0.0656**
-0.0716*
-0.105***
-0.0367
0.0500
-0.111***
(0.0277)
(0.0385)
(0.0341)
(0.0279)
(0.0380)
(0.0347)
Born in locust loc. year 79
-0.00701
-0.0139
-0.0577
-0.0120
-0.00292
-0.0663
(0.0295)
(0.0367)
(0.0395)
(0.0332)
(0.0428)
(0.0456)
-0.0228
-0.0729*
-0.0400
0.0109
0.00780
-0.0185
(0.0267)
(0.0389)
(0.0335)
(0.0290)
(0.0401)
(0.0377)
-0.0127
-0.0602
-0.0227
0.0506
0.0448
0.0295
(0.0314)
(0.0464)
(0.0408)
(0.0387)
(0.0463)
(0.0574)
Born in locust loc. year 78 Born in locust loc. year 77 Born in locust loc. year 76 Born in locust loc. year 75 Born in locust loc. year 74
0.0103
-0.0279
-0.00403
0.0123
-0.00193
0.00121
(0.0311)
(0.0464)
(0.0410)
(0.0306)
(0.0388)
(0.0413)
-0.00662
-0.0315
-0.0207
0.0262
0.0232
0.0207
(0.0323)
(0.0489)
(0.0425)
(0.0353)
(0.0415)
(0.0508)
-0.0291
-0.0864*
-0.0305
0.0246
0.0267
0.00261
(0.0322)
(0.0442)
(0.0437)
(0.0368)
(0.0475)
(0.0534)
40
Table 5 continued. (1)
(2)
(3)
(4)
(5)
(6)
Full sample Boys 0.0168
Urban Boys 0.0202
Rural Boys -0.00740
Full sample Girls 0.00105
Urban Girls 0.0270
Rural Girls -0.0257
(0.0269)
(0.0395)
(0.0342)
(0.0316)
(0.0411)
(0.0418)
0.0406
-0.00442
0.0424
-0.0230
0.0327
-0.0784*
(0.0335)
(0.0447)
(0.0460)
(0.0331)
(0.0432)
(0.0464)
0.0505
-0.0452
0.0712
-0.0173
0.0558
-0.0580
(0.0335)
(0.0429)
(0.0471)
(0.0358)
(0.0487)
(0.0491)
Born in locust loc. year 70
0.00889
-0.0663
0.0169
0.0445
0.0377
0.0316
(0.0304)
(0.0437)
(0.0392)
(0.0345)
(0.0482)
(0.0475)
Born in locust loc. year 68
0.00533
-0.000869
-0.0204
0.0481*
0.0272
0.0454
(0.0291)
(0.0408)
(0.0365)
(0.0290)
(0.0411)
(0.0379)
-0.0131
0.0480
-0.0559
0.0492
0.115**
-0.0159
(0.0341)
(0.0507)
(0.0460)
(0.0399)
(0.0519)
(0.0577)
0.0177
-0.0460
0.0241
0.0961**
-0.00328
0.124**
(0.0308)
(0.0430)
(0.0403)
(0.0376)
(0.0440)
(0.0556)
-0.0118
0.00305
-0.0348
0.0758*
0.0287
0.0781
(0.0341)
(0.0446)
(0.0479)
(0.0391)
(0.0475)
(0.0584)
0.445***
0.608***
0.427***
0.346***
0.572***
0.302***
(0.0129)
(0.0262)
(0.0143)
(0.0153)
(0.0256)
(0.0181)
yes yes yes 57 067 7 357 0.112
yes yes yes 5 465 310 0.291
yes yes yes 51 602 7 047 0.104
yes yes yes 37 735 6 307 0.069
yes yes yes 5 167 307 0.206
yes yes yes 32 568 6 000 0.059
Born in locust loc. year 73 Born in locust loc. year 72 Born in locust loc. year 71
Born in locust loc. year 67 Born in locust loc. year 66 Born in locust loc. year 65 Constant Rainfall control variables Fixed eect locality Fixed eect cohort Observations Number of localities R2 *** p<0.01, ** p<0.05, * p<0.1
Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
41
42
0.495 0.47 38
Migrants school enrollment
Non migrants school enrol.
Signif. diff. btw mig. of locust affected and non affected loc.(a) Signif. diff. btw migrants and non migrants(c)
40
1
30
.(b)
0.21
0.19
25
5.2
98 650
16 228
1
.(b)
0.32
0.33
23.2
5.2
90 286
16 061
40
40
0.06
0.31
9.4
1.7
90 774
6 751
40
6
0.094
0.43
9.8
1.7
95 681
6 534
40
.(b)
0.1
0.27
9
1.6
894 796
65 697
40
.(b)
0.18
0.42
9.3
1.53
938 251
58 248
40
40
0.03
0.19
6.8
1.9
71 525
9 661
40
0
0.07
0.3
5.1
1.8
50 626
9 286
40
.(b)
0.04
0.17
5.5
1.6
611 865
85 896
40
.(b)
0.105
0.3
4.6
1.63
492 592
87 050
Rural localities Cohorts 1980-1991 Cohorts 1965-1979 Locust affected Non affected Locust affected Non affected
(a): t-test is performed on every 40 migrants reallocation simulations. Are reported the number of significant differences, over 40 simulations, between migrants average school enrollment of locust affected and non affected villages. (b): t-test is performed over locust affected and non affected villages. t-value is reported in the locust affected column. (c): t-test is performed on every 40 migrants reallocation simulations. Are reported the number of significant diff., over 40 simulations, between average sch. enroll. of migrants and non migrants.
2
30
Signif. diff. btw mig. of locust affected and non affected loc.(a) Signif. diff. btw migrants and non migrants(c)
0.22
.(b)
0.36
Non migrants school enrol.
0.31
40.5
37
0.36
Migrants school enrollment
46.1
6.6
43 839
6 082
0.21
70.4
Average nb of non mig. per loc.
5.2
152 279
13 046
19
0
.(b) 36
0.33
0.35
34.4
6.5
36 903
5 966
0.48
0.449
47.9
4.7
158 735
11 520
0.36
5.9
62 437
Non migrants population
Average nb of migrants per loc.
4 273
Migrants population
FEMALES
76.2
Average nb of non mig. per loc.
6
5.5
63 791
Non migrants population
Average nb of migrants per loc.
3 992
Migrants population
MALES
Table A.1: Migrants characteristics and divergences between samples. Urban localities Cohorts 1980-1991 Cohorts 1965-1979 Locust affected Non affected Locust affected Non affected
43
Born in locust loc. year 75
Born in locust loc. year 76
Born in locust loc. year 77
Born in locust loc. year 78
Born in locust loc. year 79
Born in locust loc. year 80
Born in locust loc. year 81
Born in locust loc. year 82
Born in locust loc. year 83
Born in locust loc. year 84
Born in locust loc. year 85
Born in locust loc. year 86
Born in locust loc. year 87
Born in locust loc. year 88
Born in locust loc. year 89
Born in locust loc. year 90
Born in locust loc. year 91
(0.0110)
(0.0110)
(0.0101)
0.00139
0.00174
(0.0101)
(0.0112)
-0.000489
(0.0112)
-0.00113
0.0138
(0.00942)
0.0146
(0.00940)
(0.0105)
-0.00272
(0.0105)
-0.00263
0.00910
(0.00986)
(0.00984)
0.00872
0.00990
(0.00952)
0.00925
(0.00950)
(0.00988)
-5.23e-05
(0.00987)
-0.000364
0.0124
(0.00992)
0.0119
(0.00991)
(0.0111)
-0.0297***
-0.0316***
(0.0111)
(0.0106)
-0.0291***
-0.0305***
(0.0105)
(0.0110)
-0.0409***
-0.0425***
(0.0110)
(0.0114)
-0.0437***
-0.0449***
(0.0114)
(0.0106)
-0.0518***
-0.0535***
(0.0105)
(0.0108)
-0.0583***
-0.0594***
(0.0108)
(0.0103)
-0.0593***
(0.0102)
-0.0603***
-0.0539***
(0.0102)
(0.0102)
-0.0543***
-0.0178*
-0.0180*
(0.0110)
0.00115
(0.0101)
7.84e-05
(0.0112)
0.0135
(0.00943)
-0.00299
(0.0105)
0.00992
(0.00987)
0.0101
(0.00953)
-0.000383
(0.00989)
0.0126
(0.00992)
-0.0279***
(0.0111)
-0.0279**
(0.0106)
-0.0407***
(0.0110)
-0.0430***
(0.0114)
-0.0505***
(0.0106)
-0.0575***
(0.0108)
-0.0589***
(0.0103)
-0.0524***
(0.0102)
-0.0176*
(0.0110)
0.00206
(0.0102)
0.000601
(0.0112)
0.0138
(0.00945)
-0.00259
(0.0106)
0.00997
(0.00990)
0.0117
(0.00955)
0.000239
(0.00991)
0.0136
(0.00994)
-0.0266***
(0.0111)
-0.0253**
(0.0106)
-0.0379***
(0.0110)
-0.0412***
(0.0114)
-0.0489***
(0.0106)
-0.0556***
(0.0109)
-0.0567***
(0.0103)
-0.0509***
(0.0103)
-0.0153
Table A.2: Robustness test: sample year cut-off, School enrollment rate, Boys. Urban and rural localities Coh. 91 - 66 Coh. 91- 67 Coh. 91 - 68 Coh. 91 - 69
(0.0117)
-0.00270
(0.0107)
-0.00575
(0.0118)
0.00637
(0.00997)
-0.00681
(0.0112)
-0.00103
(0.0104)
0.000694
(0.00987)
-0.00762
(0.0103)
0.00446
(0.0104)
-0.0404***
(0.0115)
-0.0440***
(0.0108)
-0.0583***
(0.0115)
-0.0593***
(0.0118)
-0.0703***
(0.0110)
-0.0721***
(0.0112)
-0.0737***
(0.0108)
-0.0639***
(0.0108)
-0.0259**
Coh. 91 - 66
(0.0117)
-0.00309
(0.0107)
-0.00521
(0.0118)
0.00549
(0.00998)
-0.00701
(0.0112)
-0.000655
(0.0104)
0.00109
(0.00988)
-0.00751
(0.0103)
0.00488
(0.0104)
-0.0386***
(0.0115)
-0.0429***
(0.0109)
-0.0569***
(0.0115)
-0.0582***
(0.0118)
-0.0687***
(0.0111)
-0.0712***
(0.0112)
-0.0727***
(0.0108)
-0.0635***
(0.0108)
-0.0258**
(0.0117)
-0.00348
(0.0107)
-0.00472
(0.0118)
0.00499
(0.00999)
-0.00749
(0.0112)
-6.07e-05
(0.0104)
0.00110
(0.00990)
-0.00816
(0.0103)
0.00482
(0.0104)
-0.0369***
(0.0115)
-0.0419***
(0.0109)
-0.0570***
(0.0115)
-0.0577***
(0.0118)
-0.0676***
(0.0111)
-0.0704***
(0.0112)
-0.0725***
(0.0108)
-0.0621***
(0.0108)
-0.0256**
Rural localities Coh. 91 - 67 Coh. 91 - 68
(0.0118)
-0.00273
(0.0107)
-0.00443
(0.0119)
0.00514
(0.0100)
-0.00729
(0.0112)
-0.000230
(0.0104)
0.00242
(0.00992)
-0.00767
(0.0104)
0.00553
(0.0105)
-0.0360***
(0.0116)
-0.0394***
(0.0109)
-0.0543***
(0.0115)
-0.0564***
(0.0118)
-0.0663***
(0.0111)
-0.0688***
(0.0112)
-0.0705***
(0.0108)
-0.0609***
(0.0108)
-0.0234**
Coh. 91 - 69
44 yes yes yes 214,897 0.092 10,111
(0.00306)
(0.00305)
yes yes yes 222,485 0.090 10,111
0.131***
0.130*** yes yes yes 208,460 0.096 10,110
(0.00307)
yes yes yes 199,469 0.099 10,103
(0.00308)
yes yes yes 214,100 0.085 9,771
(0.00314)
0.123***
(0.0112)
0.132***
(0.0137)
-0.00625
(0.0105)
-0.0126
(0.0109)
-0.00936
(0.0114)
-0.0127
(0.0117)
0.00473
(0.0107)
7.90e-05
(0.0113)
-0.0103
(0.0106)
0.131***
(0.0103)
-0.0111
(0.0108)
-0.0105
(0.0109)
0.00585
(0.0100)
0.00113
(0.0106)
-0.00842
Coh. 91 - 66
-0.000765
(0.0127)
(0.00982)
-0.0120
(0.0103)
-0.0115
(0.0108)
-0.0107
(0.0109)
0.00590
(0.01000)
0.000565
(0.0106)
-0.00789
Coh. 91 - 69
-0.000218
(0.0127)
(0.00982)
-0.00177
-0.00167
(0.00981)
(0.0103)
-0.0117
-0.0119
(0.0103)
(0.0108)
-0.0116
-0.0113
(0.0108)
(0.0109)
-0.0104
(0.0109)
-0.0109
0.00568
(0.00999)
(0.00997)
0.00580
0.000824
(0.0106)
0.000464
-0.00785
(0.0106)
Urban and rural localities Coh. 91- 67 Coh. 91 - 68
-0.00753
Coh. 91 - 66
Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
*** p<0.01, ** p<0.05, * p<0.1
Rainfall control variables Fixed effect locality Fixed effect cohort Observations R-squared Number of localities
Constant
Born in locust loc. year 66
Born in locust loc. year 67
Born in locust loc. year 68
Born in locust loc. year 70
Born in locust loc. year 71
Born in locust loc. year 72
Born in locust loc. year 73
Born in locust loc. year 74
Table A.2 continued.
yes yes yes 206,816 0.088 9,771
(0.00315)
0.124***
(0.0137)
-0.00625
(0.0105)
-0.0126
(0.0109)
-0.00967
(0.0114)
-0.0123
(0.0117)
0.00452
(0.0107)
0.000257
(0.0114)
-0.0106
yes yes yes 200,679 0.092 9,770
(0.00316)
0.124***
(0.0105)
-0.0130
(0.0109)
-0.00976
(0.0114)
-0.0128
(0.0117)
0.00458
(0.0107)
-0.000108
(0.0114)
-0.0108
Rural localities Coh. 91 - 67 Coh. 91 - 68
yes yes yes 192,015 0.095 9,763
(0.00317)
0.124***
(0.0109)
-0.00948
(0.0115)
-0.0127
(0.0117)
0.00438
(0.0108)
0.000165
(0.0114)
-0.0114
Coh. 91 - 69
45
Born in locust loc. year 75
Born in locust loc. year 76
Born in locust loc. year 77
Born in locust loc. year 78
Born in locust loc. year 79
Born in locust loc. year 80
Born in locust loc. year 81
Born in locust loc. year 82
Born in locust loc. year 83
Born in locust loc. year 84
Born in locust loc. year 85
Born in locust loc. year 86
Born in locust loc. year 87
Born in locust loc. year 88
Born in locust loc. year 89
Born in locust loc. year 90
Born in locust loc. year 91
(0.151)
(0.150)
(0.144)
-0.0382
-0.0272
(0.144)
(0.144)
0.0450
0.0556
(0.143)
(0.130)
-0.0138
-0.00121
(0.130)
(0.140)
-0.0916
-0.0765
(0.139)
(0.139)
-0.110
-0.0979
(0.139)
(0.144)
-0.267*
-0.261*
(0.143)
(0.147)
-0.257*
-0.263*
(0.146)
(0.127)
-0.102
-0.109
(0.127)
(0.130)
-0.229*
-0.228*
(0.129)
(0.127)
-0.135
-0.144
(0.127)
(0.122)
-0.121
-0.120
(0.122)
(0.124)
-0.0332
-0.0429
(0.124)
(0.124)
-0.0523
-0.0575
(0.123)
(0.124)
0.00598
-0.00239
(0.124)
(0.124)
-0.0873
(0.123)
-0.0860
-0.0840
(0.126)
(0.125)
-0.0855
-0.122
-0.117
(0.150)
-0.0476
(0.144)
0.0450
(0.143)
-0.0166
(0.130)
-0.0914
(0.140)
-0.118
(0.139)
-0.273**
(0.144)
-0.256*
(0.147)
-0.103
(0.128)
-0.234*
(0.130)
-0.136
(0.127)
-0.111
(0.122)
-0.0356
(0.124)
-0.0501
(0.124)
0.00430
(0.124)
-0.0811
(0.124)
-0.0873
(0.125)
-0.128
Table A.3: Robustness test: sample year cut-off, Primary Grade, Boys. Urban and rural localities Coh. 91 - 66 Coh. 91- 67 Coh. 91 - 68
(0.150)
-0.0553
(0.144)
0.0467
(0.143)
-0.0264
(0.130)
-0.0827
(0.140)
-0.109
(0.139)
-0.268*
(0.144)
-0.255*
(0.147)
-0.106
(0.128)
-0.227*
(0.130)
-0.139
(0.127)
-0.109
(0.122)
-0.0323
(0.124)
-0.0470
(0.124)
-0.0110
(0.124)
-0.0983
(0.124)
-0.0868
(0.126)
-0.128
Coh. 91 - 69
(0.206)
0.0221
(0.199)
0.0883
(0.202)
0.0355
(0.178)
-0.0690
(0.196)
-0.232
(0.189)
-0.346*
(0.199)
-0.425**
(0.202)
-0.206
(0.173)
-0.254
(0.176)
-0.109
(0.175)
-0.138
(0.169)
-0.0139
(0.170)
-0.0263
(0.170)
0.0560
(0.171)
-0.0258
(0.170)
-0.00235
(0.171)
-0.00798
Coh. 91 - 66
(0.207)
0.00858
(0.200)
0.0773
(0.203)
0.0217
(0.179)
-0.0891
(0.197)
-0.248
(0.191)
-0.356*
(0.200)
-0.420**
(0.204)
-0.197
(0.175)
-0.256
(0.178)
-0.0982
(0.176)
-0.140
(0.170)
-0.00305
(0.172)
-0.0199
(0.171)
0.0656
(0.172)
-0.0269
(0.171)
-0.000163
(0.172)
-0.0127
(0.207)
-0.00404
(0.200)
0.0822
(0.202)
0.0192
(0.178)
-0.0926
(0.197)
-0.261
(0.191)
-0.364*
(0.201)
-0.421**
(0.204)
-0.201
(0.175)
-0.261
(0.178)
-0.0996
(0.176)
-0.132
(0.170)
-0.00794
(0.172)
-0.0190
(0.171)
0.0624
(0.172)
-0.0212
(0.171)
-0.00434
(0.172)
-0.0207
Rural localities Coh. 91 - 67 Coh. 91 - 68
(0.207)
-0.0140
(0.200)
0.0827
(0.203)
0.00670
(0.179)
-0.0806
(0.198)
-0.251
(0.191)
-0.358*
(0.201)
-0.417**
(0.204)
-0.204
(0.175)
-0.253
(0.178)
-0.105
(0.176)
-0.131
(0.170)
-0.00171
(0.172)
-0.0132
(0.172)
0.0472
(0.172)
-0.0395
(0.172)
-0.000947
(0.173)
-0.0198
Coh. 91 - 69
46 yes yes yes 77,715 0.460 7,730
(0.0424)
(0.0420)
yes yes yes 79,669 0.456 7,766
4.406***
4.412*** yes yes yes 75,981 0.464 7,707
(0.0428)
yes yes yes 73,300 0.471 7,634
(0.0436)
yes yes yes 72,987 0.438 7,449
(0.0465)
4.365***
(0.195)
4.404***
(0.210)
-0.167
(0.188)
-0.151
(0.186)
0.102
(0.197)
0.157
(0.203)
0.287
(0.188)
0.0339
(0.204)
0.0201
(0.143)
4.408***
(0.138)
-0.0127
(0.142)
0.0740
(0.145)
0.191
(0.134)
0.0155
(0.146)
-0.0376
Coh. 91 - 66
0.133
(0.152)
(0.151)
(0.139)
-0.0866
(0.138)
-0.0167
(0.142)
0.0792
(0.145)
0.193
(0.134)
0.0173
(0.145)
-0.0387
Coh. 91 - 69
0.0427
-0.108
(0.139)
-0.102
-0.0857
(0.139)
(0.138)
-0.0930
-0.0151
(0.137)
(0.142)
-0.0160
0.0764
(0.142)
(0.145)
(0.144)
0.0647
0.200
(0.134)
(0.134)
0.204
0.0185
(0.146)
0.0243
-0.0331
(0.145)
Urban and rural localities Coh. 91- 67 Coh. 91 - 68
-0.0249
Coh. 91 - 66
Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
*** p<0.01, ** p<0.05, * p<0.1
Rainfall control variables Fixed effect locality Fixed effect cohort Observations R-squared Number of localities
Constant
Born in locust loc. year 66
Born in locust loc. year 67
Born in locust loc. year 68
Born in locust loc. year 70
Born in locust loc. year 71
Born in locust loc. year 72
Born in locust loc. year 73
Born in locust loc. year 74
Table A.3 continued.
yes yes yes 71,262 0.442 7,413
(0.0469)
4.359***
(0.211)
-0.177
(0.189)
-0.143
(0.188)
0.104
(0.198)
0.174
(0.206)
0.283
(0.189)
0.0273
(0.206)
0.0108
yes yes yes 69,746 0.445 7,390
(0.0474)
4.363***
(0.189)
-0.144
(0.188)
0.102
(0.198)
0.176
(0.206)
0.276
(0.189)
0.0263
(0.206)
0.00405
Rural localities Coh. 91 - 67 Coh. 91 - 68
yes yes yes 67,309 0.451 7,318
(0.0483)
4.356***
(0.188)
0.109
(0.198)
0.167
(0.206)
0.272
(0.190)
0.0204
(0.207)
0.00367
Coh. 91 - 69
47
Born in locust loc. year 75
Born in locust loc. year 76
Born in locust loc. year 77
Born in locust loc. year 78
Born in locust loc. year 79
Born in locust loc. year 80
Born in locust loc. year 81
Born in locust loc. year 82
Born in locust loc. year 83
Born in locust loc. year 84
Born in locust loc. year 85
Born in locust loc. year 86
Born in locust loc. year 87
Born in locust loc. year 88
Born in locust loc. year 89
Born in locust loc. year 90
-0.00269 (0.00589)
(0.00587)
(0.00566)
(0.00565)
-0.00247
0.00556
(0.00653)
(0.00651)
0.00515
-0.000526
(0.00531)
(0.00530)
0.000373
-0.000303
-0.000195
0.00985 (0.00636)
0.0100
(0.00569)
(0.00566) (0.00635)
0.00991*
0.00923
0.00348 (0.00583)
0.00339
(0.00571)
(0.00569) (0.00582)
0.00779
(0.00595)
0.00780
(0.00594)
(0.00680)
-0.00775
(0.00679)
-0.00889
-0.0108
(0.00745)
-0.0116*
(0.00744)
(0.00721)
-0.00754
-0.00819
(0.00719)
(0.00833)
-0.0234***
-0.0237***
(0.00832)
(0.00703)
-0.0252***
-0.0256***
(0.00701)
(0.00785)
-0.0341***
-0.0346***
(0.00783)
(0.00714)
-0.0331***
(0.00712)
-0.0335***
-0.0349***
-0.0347***
(0.00642)
(0.00590)
-0.00324
(0.00568)
0.00551
(0.00654)
-0.000712
(0.00533)
-0.000787
(0.00636)
0.0103
(0.00569)
0.00984*
(0.00584)
0.00390
(0.00571)
0.00798
(0.00595)
-0.00697
(0.00681)
-0.00961
(0.00745)
-0.00687
(0.00722)
-0.0224***
(0.00835)
-0.0247***
(0.00704)
-0.0332***
(0.00786)
-0.0329***
(0.00714)
-0.0341***
(0.00642)
-0.0190***
(0.00592)
-0.00250
(0.00570)
0.00592
(0.00657)
-0.000417
(0.00535)
-0.000229
(0.00639)
0.00981
(0.00571)
0.0107*
(0.00586)
0.00445
(0.00574)
0.00882
(0.00597)
-0.00623
(0.00686)
-0.00800
(0.00745)
-0.00495
(0.00724)
-0.0209***
(0.00836)
-0.0233***
(0.00707)
-0.0319***
(0.00789)
-0.0313***
(0.00717)
-0.0333***
(0.00644)
-0.0171***
(0.00616)
-0.00529
(0.00596)
0.00409
(0.00683)
-0.00475
(0.00550)
-0.00138
(0.00639)
0.00312
(0.00588)
0.00611
(0.00606)
-0.00178
(0.00585)
0.00401
(0.00614)
-0.0160***
(0.00694)
-0.0225***
(0.00761)
-0.0208***
(0.00731)
-0.0354***
(0.00835)
-0.0445***
(0.00713)
-0.0458***
(0.00787)
-0.0498***
(0.00715)
-0.0480***
(0.00660)
-0.0270***
-0.0186***
(0.00640)
Born in locust loc. year 91
-0.0188***
Coh. 91 - 66
Table A.4: Robustness test: sample year cut-off, School enrollment rate, Girls. Urban and rural localities Coh. 91 - 66 Coh. 91- 67 Coh. 91 - 68 Coh. 91 - 69
(0.00616)
-0.00556
(0.00597)
0.00437
(0.00685)
-0.00578
(0.00551)
-0.00158
(0.00640)
0.00290
(0.00590)
0.00656
(0.00607)
-0.00190
(0.00588)
0.00393
(0.00615)
-0.0150**
(0.00694)
-0.0219***
(0.00762)
-0.0203***
(0.00732)
-0.0352***
(0.00836)
-0.0443***
(0.00715)
-0.0456***
(0.00788)
-0.0495***
(0.00716)
-0.0481***
(0.00661)
-0.0269***
(0.00618)
-0.00615
(0.00598)
0.00426
(0.00687)
-0.00594
(0.00553)
-0.00214
(0.00641)
0.00327
(0.00590)
0.00645
(0.00608)
-0.00158
(0.00588)
0.00404
(0.00615)
-0.0143**
(0.00695)
-0.0208***
(0.00762)
-0.0198***
(0.00733)
-0.0343***
(0.00837)
-0.0439***
(0.00716)
-0.0448***
(0.00789)
-0.0494***
(0.00716)
-0.0474***
(0.00662)
-0.0272***
Rural localities Coh. 91 - 67 Coh. 91 - 68
(0.00621)
-0.00558
(0.00601)
0.00445
(0.00690)
-0.00580
(0.00555)
-0.00181
(0.00644)
0.00262
(0.00592)
0.00709
(0.00609)
-0.00114
(0.00590)
0.00453
(0.00616)
-0.0139**
(0.00700)
-0.0194***
(0.00762)
-0.0181**
(0.00734)
-0.0332***
(0.00838)
-0.0427***
(0.00718)
-0.0438***
(0.00792)
-0.0481***
(0.00719)
-0.0469***
(0.00664)
-0.0255***
Coh. 91 - 69
48
(0.00888)
0.0137
(0.00581)
0.00198
(0.00588)
-0.00532
(0.00630)
-0.00402
(0.00621)
-0.00675
(0.00563)
-0.00216
yes yes yes 221,151 0.100 10,111
(0.00193)
(0.00191)
yes yes yes 229,513 0.099 10,112
0.0505***
0.0503*** yes yes yes 214,456 0.103 10,111
(0.00194)
yes yes yes 204,947 0.103 10,109
(0.00196)
yes yes yes 221,057 0.091 9,772
(0.00194)
0.0434***
(0.00665)
0.0510***
(0.00563)
-0.00565
(0.00610)
-0.00479
(0.00596)
-0.00405
(0.00539)
-0.00190
(0.00687)
0.000985
(0.00630)
0.0504***
(0.00561)
0.000711
(0.00562)
-0.00616
(0.00609)
-0.00537
(0.00594)
-0.00422
(0.00537)
-0.00277
0.00343 (0.00667)
Coh. 91 - 66
0.00160
(0.00819)
(0.00818)
0.00345 (0.00665)
Coh. 91 - 69
-0.000304
0.0123
(0.00560)
(0.00558)
0.0127
0.00111
(0.00560)
0.00112
(0.00560)
(0.00608)
-0.00598
-0.00571
(0.00607)
(0.00591)
-0.00490
-0.00495
(0.00590)
(0.00536)
-0.00438
(0.00534)
-0.00416
-0.00210
-0.00243
0.00340 (0.00664)
0.00384
Urban and rural localities Coh. 91- 67 Coh. 91 - 68
(0.00663)
Coh. 91 - 66
Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
*** p<0.01, ** p<0.05, * p<0.1
Rainfall control variables Fixed effect locality Fixed effect cohort Observations R-squared Number of localities
Constant
Born in locust loc. year 66
Born in locust loc. year 67
Born in locust loc. year 68
Born in locust loc. year 70
Born in locust loc. year 71
Born in locust loc. year 72
Born in locust loc. year 73
Born in locust loc. year 74
Table A.4 continued.
yes yes yes 213,012 0.092 9,771
(0.00196)
0.0437***
(0.00889)
0.0133
(0.00582)
0.00181
(0.00587)
-0.00570
(0.00631)
-0.00416
(0.00623)
-0.00714
(0.00564)
-0.00202
(0.00687)
0.000421
yes yes yes 206,618 0.095 9,771
(0.00196)
0.0436***
(0.00583)
0.00139
(0.00589)
-0.00590
(0.00632)
-0.00465
(0.00625)
-0.00704
(0.00566)
-0.00275
(0.00688)
0.000419
Rural localities Coh. 91 - 67 Coh. 91 - 68
yes yes yes 197,437 0.095 9,769
(0.00199)
0.0442***
(0.00591)
-0.00559
(0.00634)
-0.00424
(0.00628)
-0.00708
(0.00568)
-0.00217
(0.00690)
0.000238
Coh. 91 - 69
49
Born in locust loc. year 75
Born in locust loc. year 76
Born in locust loc. year 77
Born in locust loc. year 78
Born in locust loc. year 79
Born in locust loc. year 80
Born in locust loc. year 81
Born in locust loc. year 82
Born in locust loc. year 83
Born in locust loc. year 84
Born in locust loc. year 85
Born in locust loc. year 86
Born in locust loc. year 87
Born in locust loc. year 88
Born in locust loc. year 89
Born in locust loc. year 90
Born in locust loc. year 91
(0.172)
(0.172)
(0.159)
-0.219
-0.211
(0.159)
(0.182)
-0.148
-0.142
(0.182)
(0.158)
-0.0236
-0.0176
(0.157)
(0.148)
-0.195
-0.198
(0.148)
(0.155)
-0.215
-0.218
(0.155)
(0.167)
-0.318**
-0.308**
(0.167)
(0.167)
-0.488***
-0.488***
(0.166)
(0.147)
-0.370**
-0.374**
(0.147)
(0.153)
-0.250*
-0.253*
(0.152)
(0.143)
-0.361**
-0.354**
(0.143)
(0.137)
-0.263*
-0.262*
(0.137)
(0.144)
-0.203
-0.202
(0.144)
(0.135)
-0.136
-0.128
(0.135)
(0.137)
-0.238*
-0.235*
(0.136)
(0.139)
-0.193
(0.139)
-0.187
-0.265*
(0.140)
(0.140)
-0.265*
-0.369***
-0.364***
(0.173)
-0.221
(0.160)
-0.153
(0.183)
-0.0267
(0.159)
-0.202
(0.149)
-0.222
(0.156)
-0.318**
(0.167)
-0.490***
(0.168)
-0.368**
(0.147)
-0.251*
(0.153)
-0.365**
(0.144)
-0.260*
(0.138)
-0.205
(0.145)
-0.135
(0.136)
-0.240*
(0.137)
-0.191
(0.140)
-0.269*
(0.141)
-0.369***
Table A.5: Robustness test: sample year cut-off, Primary Grade,Girls. Urban and rural localities Coh. 91 - 66 Coh. 91- 67 Coh. 91 - 68
(0.174)
-0.232
(0.160)
-0.149
(0.182)
-0.0127
(0.158)
-0.204
(0.150)
-0.239
(0.157)
-0.318**
(0.168)
-0.490***
(0.168)
-0.377**
(0.147)
-0.255*
(0.153)
-0.359**
(0.144)
-0.254*
(0.138)
-0.203
(0.145)
-0.138
(0.136)
-0.242*
(0.138)
-0.201
(0.140)
-0.256*
(0.141)
-0.365***
Coh. 91 - 69
(0.245)
-0.425*
(0.227)
-0.367
(0.256)
-0.305
(0.222)
-0.465**
(0.209)
-0.480**
(0.212)
-0.611***
(0.223)
-1.031***
(0.225)
-0.793***
(0.195)
-0.576***
(0.201)
-0.638***
(0.199)
-0.578***
(0.190)
-0.426**
(0.199)
-0.318
(0.185)
-0.441**
(0.190)
-0.336*
(0.193)
-0.420**
(0.193)
-0.524***
Coh. 91 - 66
(0.245)
-0.441*
(0.227)
-0.384*
(0.257)
-0.320
(0.223)
-0.473**
(0.209)
-0.484**
(0.212)
-0.631***
(0.223)
-1.035***
(0.226)
-0.796***
(0.195)
-0.580***
(0.201)
-0.652***
(0.199)
-0.588***
(0.190)
-0.435**
(0.199)
-0.334*
(0.185)
-0.449**
(0.190)
-0.351*
(0.193)
-0.428**
(0.193)
-0.537***
(0.247)
-0.445*
(0.229)
-0.393*
(0.258)
-0.324
(0.225)
-0.481**
(0.211)
-0.489**
(0.214)
-0.629***
(0.224)
-1.035***
(0.228)
-0.789***
(0.197)
-0.578***
(0.203)
-0.656***
(0.201)
-0.583***
(0.192)
-0.436**
(0.201)
-0.329
(0.187)
-0.450**
(0.192)
-0.345*
(0.195)
-0.431**
(0.195)
-0.534***
Rural localities Coh. 91 - 67 Coh. 91 - 68
(0.248)
-0.459*
(0.231)
-0.387*
(0.259)
-0.304
(0.225)
-0.483**
(0.213)
-0.513**
(0.216)
-0.627***
(0.225)
-1.035***
(0.229)
-0.801***
(0.198)
-0.584***
(0.203)
-0.650***
(0.201)
-0.573***
(0.192)
-0.428**
(0.201)
-0.330
(0.188)
-0.447**
(0.193)
-0.351*
(0.196)
-0.407**
(0.196)
-0.525***
Coh. 91 - 69
50 yes yes yes 54,390 0.395 6,911
(0.0535)
(0.0530)
yes yes yes 55,665 0.388 6,951
4.023***
4.029*** yes yes yes 53,409 0.398 6,886
(0.0540)
Standard errors corrected for clustering and auto-correlation by clustering at the village level.
yes yes yes 51,500 0.404 6,802
(0.0546)
yes yes yes 49,317 0.365 6,637
(0.0623)
3.906***
(0.239)
4.014***
(0.242)
-0.450*
(0.208)
-0.362*
(0.243)
-0.394
(0.235)
-0.331
(0.244)
-0.499**
(0.215)
-0.434**
(0.239)
-0.225
(0.161)
4.016***
(0.173)
-0.137
(0.169)
-0.114
(0.173)
-0.230
(0.159)
-0.187
(0.171)
-0.0581
Coh. 91 - 66
0.127
(0.174)
(0.152)
-0.195
(0.173)
-0.144
(0.169)
-0.120
(0.173)
-0.238
(0.158)
-0.200
(0.171)
-0.0815
Coh. 91 - 69
0.173
(0.174)
(0.151)
-0.147
-0.147
(0.151)
(0.172)
-0.196
-0.204
(0.172)
(0.168)
-0.145
-0.147
(0.168)
(0.173)
-0.124
-0.125
(0.173)
(0.157)
-0.240
(0.157)
-0.238
-0.199
(0.171)
-0.198
-0.0781
(0.170)
Urban and rural localities Coh. 91- 67 Coh. 91 - 68
-0.0709
Coh. 91 - 66
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities.
Rainfall control variables Fixed effect locality Fixed effect cohort Observations R-squared Number of localities
Constant
Born in locust loc. year 66
Born in locust loc. year 67
Born in locust loc. year 68
Born in locust loc. year 70
Born in locust loc. year 71
Born in locust loc. year 72
Born in locust loc. year 73
Born in locust loc. year 74
Table A.5 continued.
yes yes yes 48,258 0.371 6,597
(0.0627)
3.896***
(0.242)
-0.456*
(0.208)
-0.356*
(0.244)
-0.397
(0.236)
-0.334
(0.245)
-0.503**
(0.215)
-0.441**
(0.239)
-0.242
yes yes yes 47,470 0.375 6,572
(0.0632)
3.888***
(0.210)
-0.354*
(0.246)
-0.393
(0.237)
-0.330
(0.246)
-0.502**
(0.217)
-0.446**
(0.240)
-0.247
Rural localities Coh. 91 - 67 Coh. 91 - 68
yes yes yes 45,804 0.380 6,490
(0.0639)
3.884***
(0.247)
-0.378
(0.238)
-0.323
(0.246)
-0.492**
(0.219)
-0.431**
(0.240)
-0.213
Coh. 91 - 69
Table A.6 : Impact of locust invasion on education without rainfall variables, rural localities. (1) (2) (3) (4) VARIABLES Boys Boys Girls Girls School enrollment Grade School enrollment Grade Born in locust loc. year 91 -0.0277*** -0.00415 -0.0283*** -0.497*** (0.0107)
(0.170)
(0.00652)
(0.190)
Born in locust loc. year 90
-0.0682***
0.0169
-0.0505***
-0.401**
(0.0107)
(0.169)
(0.00711)
(0.189)
Born in locust loc. year 89
-0.0873***
-0.00526
-0.0598***
-0.335*
(0.0111)
(0.170)
(0.00782)
(0.187)
Born in locust loc. year 88
-0.0902***
0.0658
-0.0556***
-0.410**
(0.0109)
(0.169)
(0.00706)
(0.182)
Born in locust loc. year 87
-0.0819***
-0.0514
-0.0496***
-0.346*
(0.0117)
(0.168)
(0.00828)
(0.195)
Born in locust loc. year 86
-0.0707***
-0.0125
-0.0406***
-0.437**
(0.0114)
(0.167)
(0.00723)
(0.187)
Born in locust loc. year 85
-0.0691***
-0.152
-0.0273***
-0.568***
(0.0107)
(0.172)
(0.00757)
(0.196)
-0.0557***
-0.147
-0.0286***
-0.641***
(0.0114)
(0.175)
(0.00687)
(0.197)
-0.0442***
-0.294*
-0.0155**
-0.601***
(0.0104)
(0.171)
(0.00608)
(0.192)
0.00286
-0.202
0.00500
-0.803***
(0.0103)
(0.199)
(0.00581)
(0.222)
-0.00366
-0.465**
0.00127
-1.031***
(0.00978)
(0.196)
(0.00602)
(0.219)
0.00566
-0.363*
0.00900
-0.592***
(0.0103)
(0.187)
(0.00580)
(0.208)
-0.00175
-0.236
0.00378
-0.431**
(0.0111)
(0.193)
(0.00633)
(0.205)
-0.00203
-0.0773
0.00227
-0.423*
(0.00994)
(0.175)
(0.00547)
(0.217)
0.0101
-0.00461
-0.00259
-0.323
(0.0118)
(0.201)
(0.00676)
(0.251)
Born in locust loc. year 84 Born in locust loc. year 83 Born in locust loc. year 82 Born in locust loc. year 81 Born in locust loc. year 80 Born in locust loc. year 79 Born in locust loc. year 78 Born in locust loc. year 77
51
Table A.6 continued. (1) Boys School enrollment -0.000808
(2) Boys Grade 0.0848
(3) Girls School enrollment 0.00696
(4) Girls Grade -0.347
(0.0106)
(0.197)
(0.00591)
(0.225)
Born in locust loc. year 75
-0.000833
0.000789
-0.00440
-0.394
(0.0117)
(0.205)
(0.00609)
(0.242)
Born in locust loc. year 74
-0.00636
0.00662
0.00411
-0.202
(0.0113)
(0.205)
(0.00683)
(0.235)
Born in locust loc. year 73
0.00612
-0.00250
0.00147
-0.434**
(0.0107)
(0.188)
(0.00557)
(0.209)
Born in locust loc. year 72
0.00317
0.275
-0.00731
-0.462*
(0.0116)
(0.204)
(0.00620)
(0.241)
Born in locust loc. year 71
-0.00770
0.155
0.000553
-0.307
(0.0114)
(0.196)
(0.00628)
(0.233)
Born in locust loc. year 70
-0.00677
0.0901
-0.00348
-0.394
(0.0108)
(0.186)
(0.00583)
(0.241)
-0.0136
-0.155
0.00171
-0.318
(0.0105)
(0.188)
(0.00577)
(0.206)
-0.00573
-0.145
0.0148*
-0.444*
(0.0137)
(0.211)
(0.00886)
(0.241)
VARIABLES Born in locust loc. year 76
Born in locust loc. year 68 Born in locust loc. year 67 Born in locust loc. year 66 Born in locust loc. year 65 Constant Rainfall control variables Fixed eect locality Fixed eect cohort Observations Number of localities R2
0.00625
0.142
0.00742
0.145
(0.0112)
(0.194)
(0.00664)
(0.237)
0.00228
-0.178
0.0213**
-0.154
(0.0124)
(0.228)
(0.00905)
(0.255)
0.115***
4.310***
0.0404***
3.869***
(0.00286)
(0.0370)
(0.00175)
(0.0522)
no yes yes 220 684 9 771 0.080
no yes yes 74 540 7 480 0.433
no yes yes 227 963 9 772 0.088
no yes yes 50 099 6 652 0.362
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses. Cohort of reference: 1969. Observations correspond to number of Cohorts times number of localities. Standard errors corrected for clustering and auto-correlation by clustering at the village level.
52